342 tools. One daemon.

Native tools are served by chittad, a C++ daemon reached over a Unix socket. 55 are listed to the model by default; the remaining 287 stay callable through the advanced gateway, which keeps the default tool list small. Composed gateways run in chitta-mcp. Click a row for its parameters; additional native CLI entries appear below.

Status as of 2026-09-20 · Generated from frozen contracts on 2026-09-17, 342 tools — regenerate with python3 scripts/gen-tools-static.py --docs

Core Memory

53 tools ▾

ack_memoryIncrement ack signal for a memory. Records [ack] memory:<id> score:+1 with tag ack-signal.gateway, via advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID to ack

approve_memoryApprove a Proposed memory, promoting it to Activevia advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID to approve

create_episodeCreate dialogue episode for conversation trackingvia advanced

Parameter	Type	Required	Default
end_turn	any	No	—
episode_type	string	No	—
realm	string	No	—
session_id	string	Yes	—
start_turn	any	Yes	—
title	string	Yes	—

episode_cluster_statusFind similar episode clustersvia advanced

Parameter	Type	Required	Default	Description
min_occurrences	any	No	—
similarity_threshold	number	No	—

expand_memoryExpand a memory to full hierarchical contextvia advanced

Parameter	Type	Required	Default	Description
depth	any	No	—
id	string	Yes	—

forgetRemove a memory

Parameter	Type	Required	Default	Description
id	string	Yes	—	Node ID to forget

forget_kindBulk-delete all memories of a given kind (e.g. 'habit'). Optionally filter by realm.via advanced

Parameter	Type	Required	Default	Description
kind	string	Yes	—	Memory kind to delete (e.g. habit, unknown)
limit	any	No	—	Max to delete (default 5000)
realm	string	No	—	Optional realm filter

getGet a node by ID

Parameter	Type	Required	Default	Description
id	string	Yes	—

get_entitiesGet tracked entitiesvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—
type	string	No	—

get_evidence_typeRetrieve the evidence type tag of a memoryvia advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID

labile_memoriesList memories recalled multiple times recently — candidates for reconsolidation (excludes freshly-written hook memories)via advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max results (default 20)
min_access	any	No	—	Min recall count to qualify (default 2)
realm	string	No	—	Filter by realm
window_hours	number	No	—	Recency window in hours (default 48)

labile_memories_topList the most-accessed (most labile) memories — candidates for reconsolidationvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max results (default 20)
realm	string	No	—	Filter by realm

list_aspectsList available semantic aspectsvia advanced

list_by_aspectList memories by semantic aspectvia advanced

Parameter	Type	Required	Default
aspect	string	Yes	—
limit	any	No	—
min_confidence	number	No	—

list_by_statusList memories filtered by lifecycle status (active/superseded/contradicted/archived)via advanced

Parameter	Type	Required	Default	Description
limit	any	No	50	Max results
realm	string	No	—	Filter by realm
status	string	No	"superseded"	Filter: active, superseded, contradicted, archived, or all

list_memories_briefFast memory indexvia advanced

Parameter	Type	Required	Default
kind	string	No	—
limit	any	No	—
priority_tier	any	No	—
realm	string	No	—

list_merge_queueList pending merge proposalsvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—
status	string	No	—

list_pinnedList pinned memoriesvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—
realm	string	No	—

lookupUnified memory lookup. Classifies query intent, fans out to optimal backends (keyword/semantic/triplet/temporal/code), fuses with weighted RRF. Default entry point for memory search — use instead of recall/smart_recall/hybrid_recall.gateway

Parameter	Type	Required	Default	Description
explain	boolean	No	—	Include intent classification and score breakdown
limit	integer	No	—	Max results (default: 10)
mode	string (auto\|fast\|deep)	No	—	auto=escalate if low confidence; fast=skip deep search; deep=force full resonate
query	string	Yes	—	Search query
realm	string	No	—	Filter by realm

mark_memory_invalidatedMark a memory as invalidated by a commit hash or symbol-change ID.via advanced

Parameter	Type	Required	Default	Description
memory_id	any	Yes	—	Memory ID to mark
reason	string	No	—	Commit hash or change ID causing invalidation

memory_editEdit memory metadata. Actions: set_type (change memory type), set_priority (change priority tier)gateway

Parameter	Type	Required	Default	Description
action	string	Yes	—	set_type\|set_priority
id	any	Yes	—	Memory ID
tier	string	No	—	New priority tier (for set_priority)
type	string	No	—	New type (for set_type)

memory_historyView memory version historyvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—
limit	any	No	—

memory_lockAcquire memory lockvia advanced

Parameter	Type	Required	Default
duration	any	No	—
holder_id	string	Yes	—
holder_type	string	No	—
id	any	Yes	—

memory_lock_statusCheck lock statusvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—

memory_outcomeRecord an observed outcome for a memory into its Beta utility posterior (success raises alpha, failure raises beta) Routed by the chitta-mcp gateway before it reaches the daemon, so the effective arguments can differ from the daemon schema below.gateway

Parameter	Type	Required	Default	Description
id	string	Yes	—	Node ID the outcome is about
success	boolean	Yes	—	Did acting on this memory work?
weight	number	No	—	Observation weight, capped at 5.0 (default 1.0)

memory_provenanceShow why a memory exists: source, evidence, superseded_by, supersedes relationsvia advanced

Parameter	Type	Required	Default	Description
id	string	No	—	Memory ID to inspect
memory_id	any	No	—	Memory ID (numeric)

memory_revertRevert memory to previous versionvia advanced

Parameter	Type	Required	Default
id	any	Yes	—
reason	string	No	—
version	any	Yes	—

memory_statusGet effective status of a memory: active, superseded, or contradicted — checks incoming supersedes tripletsvia advanced

Parameter	Type	Required	Default	Description
id	string	No	—
memory_id	any	No	—

memory_type_statsGet memory type statisticsvia advanced

Parameter	Type	Required	Default	Description
realm	string	No	—

memory_unlockRelease memory lockvia advanced

Parameter	Type	Required	Default	Description
holder_id	string	Yes	—
id	any	Yes	—

nack_memoryDecrement nack signal for a memory. Records [nack] memory:<id> score:-1 with tag nack-signal.gateway, via advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID to nack

pin_memoryPin memory to keep hotvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—
reason	string	No	—

promote_memoryPromote a memory one tier: Proposed→Observed→Verified→Activevia advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID

propose_changePropose change to memoryvia advanced

Parameter	Type	Required	Default
content	string	Yes	—
id	any	Yes	—
proposed_by	string	Yes	—

prune_episodesPrune old/excess episode memories. Deletes episodes older than max_age_days (strength<0.3) and caps total count at max_count.via advanced

Parameter	Type	Required	Default	Description
max_age_days	any	No	—	Delete episodes older than this (default 90)
max_count	any	No	—	Cap total episode count at this (default 10000)

recallSearch memory by semantic similarity with realm filtering Routed by the chitta-mcp gateway before it reaches the daemon, so the effective arguments can differ from the daemon schema below.gateway

Parameter	Type	Required	Default	Description
explain	boolean	No	—	Include score decomposition per hit (default: False)
gwt_mode	boolean	No	—	Global Workspace Theory mode (default: False)
include_global	boolean	No	—	Include global memories (default: True)
limit	any	No	—	Max results (default 10)
min_confidence	number	No	—	Minimum confidence threshold
query	string	Yes	—	Search query
realm	string	No	—	Filter by realm
separation_mode	boolean	No	—	Diverse results via MMR (default: False)
sources	boolean	No	true	Include indexed repository sources; false returns memory ranking only (default: True)
tag	string	No	—	Filter by tag

reject_memoryReject a Proposed memory, archiving itvia advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID to reject

rememberStore text in memory with optional tags and realm

Parameter	Type	Required	Default	Description
confidence	number	No	—	Initial confidence 0-1 (default: 0.8)
content	string	Yes	—	Text to remember
realm	string	No	—	Primary realm (default: brahman)
shared_realms	array<string>	No	—	Additional realms
tags	array<string>	No	—	Optional tags
type	string	No	—	Node type (wisdom, insight, signal, episode)
visibility	any	No	—	0=Private, 1=Shared, 2=Global (default: 0)

remember_batchStore multiple memories in a single round-trip (high-throughput bulk ingest)via advanced

Parameter	Type	Required	Default	Description
items	array<object>	Yes	—	Array of memory objects (same fields as remember)
realm	string	No	—	Default realm for all items

remember_typedStore a typed memory node (digest-node, decision, open-question, etc.) with optional graph links (supersedes, invalidated-by, anchors-to).gateway, via advanced

Parameter	Type	Required	Default	Description
content	string	Yes	—	SSL triplet content for the memory
links	array<object>	No	—	Graph links to other memories
node_type	string (digest-node\|symbol-summary\|decision\|open-question\|rollup\|working-brief)	Yes	—	Type of memory node
realm	string	No	—	Realm to store in (default: brahman)
subject	string	Yes	—	What this memory is about

resolve_mergeResolve merge proposalvia advanced

Parameter	Type	Required	Default
merge_id	any	Yes	—
resolution	string	No	—
status	string	Yes	—

set_affectSet affect dimensions (valence, arousal) on a memoryvia advanced

Parameter	Type	Required	Default	Description
arousal	number	Yes	—	Emotional arousal: 0.0 to 1.0
id	string	Yes	—
valence	number	Yes	—	Emotional valence: -1.0 to +1.0

set_evidence_typeTag a memory with its epistemological evidence class (observation/inference/hearsay/authoritative/prediction)via advanced

Parameter	Type	Required	Default	Description
evidence_type	string	Yes	—	One of: observation, inference, hearsay, authoritative, prediction
id	string	Yes	—	Memory ID

set_memory_typeSet memory semantic type

Parameter	Type	Required	Default	Description
memory_id	any	Yes	—
type	string	Yes	—

set_priority_tierSet memory priority tiervia advanced

Parameter	Type	Required	Default	Description
memory_id	any	Yes	—
tier	any	Yes	—

strengthenIncrease confidence of a memoryvia advanced

Parameter	Type	Required	Default	Description
amount	number	No	—	Amount (default 0.1)
id	string	Yes	—	Node ID

tagAdd or remove tags from a nodevia advanced

Parameter	Type	Required	Default
add	string	No	—
id	string	Yes	—
remove	string	No	—

unpin_memoryUnpin a memoryvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—

updateUpdate node contentvia advanced

Parameter	Type	Required	Default	Description
content	string	Yes	—
id	string	Yes	—

verify_correctionMark a correction memory as verified at a specific code locus. Stores a verification record and updates the original correction's tags.gateway, via advanced

Parameter	Type	Required	Default	Description
evidence_locus	string	Yes	—	Location of evidence, e.g. 'sadhana_manager.cpp:299'
id	any	Yes	—	Memory/triplet ID of the correction to verify

weakenDecrease confidence of a memoryvia advanced

Parameter	Type	Required	Default	Description
amount	number	No	—
id	string	Yes	—

what_do_i_know_aboutIntrospection: return claims + provenance + staleness + contradictions for a topicvia advanced

Parameter	Type	Required	Default	Description
confidence_min	number	No	—	Minimum confidence threshold (default 0.0)
include_contradictions	boolean	No	—	Include contradiction info (default true)
include_stale	boolean	No	—	Include stale memories (default true)
k	any	No	—	Max claims to return (default 10)
realm	string	No	—	Realm filter
topic	string	Yes	—	Topic or question to introspect

write_gate_statsShow write-gate admission stats: staged memory count and oldest staged memory agevia advanced

Recall & Search

26 tools ▾

5w_searchMulti-dimensional semantic search across who/what/when/where/why axesvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max results (default 10)
realm	string	No	—	Filter by realm
what	string	No	—	What is happening/topic
when	string	No	—	Temporal description
where	string	No	—	Location or context
who	string	No	—	Who is involved
why	string	No	—	Motivation or reason

expand_queryExpand query into typed variantsvia advanced

Parameter	Type	Required	Default	Description
query	string	Yes	—

full_resonateSemantic search with full contextvia advanced

Parameter	Type	Required	Default
exclude_kinds	array<string>	No	—
include_global	boolean	No	—
k	any	No	—
partnership_only	boolean	No	—
query	string	Yes	—
realm	string	No	—
separation_mode	boolean	No	—

hybrid_recallCombined vector + BM25 + graph recallvia advanced

Parameter	Type	Required	Default	Description
bm25_weight	number	No	—
explain	boolean	No	—	Include score decomposition per hit (default: False)
graph_weight	number	No	—
limit	any	No	—
query	string	Yes	—
realm	string	No	—
recency_weight	number	No	—
tag	string	No	—
vector_weight	number	No	—

query_claimsQuery semantic claimsvia advanced

Parameter	Type	Required	Default
active_only	boolean	No	—
limit	any	No	—
predicate	string	No	—
scope	string	No	—
subject	string	No	—

recall_analogyAnalogical recall over the triplet lane (vector-symbolic, no LLM/GPU). Two modes embedding similarity cannot express, because both are about structure rather than wording. 'proportional' solves a:b :: c:? — give three entities, get ranked fillers for the fourth. 'structural' finds memories whose relation graph has the same SHAPE as a probe memory, with entity names factored out, so a pattern learned in one project matches the same pattern in another; set exclude_realm (or cross_realm) for that cross-project transfer. Use when you want 'what else looks like this?', not 'what mentions these words?'.

Parameter	Type	Required	Default	Description
a	string	No	—	proportional: source entity
b	string	No	—	proportional: what a maps to
c	string	No	—	proportional: target entity; the tool returns its counterpart to b
cross_realm	boolean	No	—	structural: shorthand for excluding the probe memory's own realm (default false)
exclude_realm	string	No	—	structural: drop results from this realm — the cross-project transfer case
limit	any	No	—	Max results, 1-100 (default 8)
memory_id	any	No	—	structural: probe memory whose relation shape is matched (excluded from results)
mode	string	No	—	'proportional' (a:b :: c:?) or 'structural' (shape match). Default: structural
realm	string	No	—	structural: restrict results to this realm
text	string	No	—	structural: free-text probe, used when memory_id is absent; anchors on the entities it mentions

recall_by_priorityBudget-aware recallvia advanced

Parameter	Type	Required	Default
budget_tokens	any	No	—
include_global	boolean	No	—
query	string	No	—
realm	string	No	—

recall_causal_antecedentPMI-ranked causal antecedents: what actions typically precede (tool, entity)?via advanced

Parameter	Type	Required	Default
entity	string	Yes	—
k	any	No	—
tool	string	Yes	—

recall_counterfactualCDAWG sibling-edge counterfactual: what alternative tool/entity would have had a lower failure rate in this same context?via advanced

Parameter	Type	Required	Default	Description
entity	string	Yes	—
k	any	No	—
outcome	any	No	—	0=success 1=fail 2=error 3=partial (default 1)
tool	string	Yes	—

recall_failure_patternReturn top-k CDAWG states with high failure rates (fail_ratio > 0.6, fail_count >= 3)via advanced

Parameter	Type	Required	Default	Description
k	any	No	—

recall_hdcbindHeteroassociative HDC query: given known_role=known_val, infer query_role. Roles: tool, entity, outcome.via advanced

Parameter	Type	Required	Default
k	any	No	—
known_role	string	Yes	—
known_val	string	Yes	—
query_role	string	Yes	—

recall_keywordBM25 keyword search

Parameter	Type	Required	Default	Description
explain	boolean	No	—	Include score decomposition per hit (default: False)
limit	any	No	—	Max results (default 10)
query	string	Yes	—	Search query

recall_lanesFan-in prompt recall lanes over one daemon RPC

Parameter	Type	Required	Default	Description
budget_ms	any	No	—	Per-lane wall-clock budget override; capped by CHITTA_RPC_BUDGET_MS
ctx_query	string	No	—	Optional context query; defaults to query
lanes	array<string>	No	—	Subset of sem, ctx, hyb, kw, corr, corrk (default all)
limits	object	No	—	Per-lane result limits (sem, ctx, hyb, kw, corr)
query	string	Yes	—	Prompt query shared by sem, hyb, kw, corr, and corrk
realm	string	No	—	Realm for sem, ctx, hyb, and kw

recall_last_actionReturn last k occurrences of (tool, entity) from the CEC event tapevia advanced

Parameter	Type	Required	Default
entity	string	Yes	—
k	any	No	—
tool	string	Yes	—

recall_motif_valueReturn top-k CDAWG motif states reachable from (tool, entity) ranked by Q-value: which action sequences have the highest expected success rate from this context?via advanced

Parameter	Type	Required	Default	Description
entity	string	Yes	—
k	any	No	—	Max states to return (default 5)
tool	string	Yes	—

recall_sessionSession-level recall: groups chunk evidence by source session using noisy-OR aggregation. Returns ranked sessions with best evidence.via advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max sessions to return (default 10)
query	string	Yes	—	Natural language query
realm	string	No	—	Realm filter (optional)

recall_smartMulti-lane retrieval planner: uses a fast LLM call to extract entities and speech-act type, then fans out to semantic, typed, spreading-activation, and session-level lanes, merging results with Reciprocal Rank Fusion. Best for complex queries.gateway, via advanced

Parameter	Type	Required	Default	Description
limit	integer	No	10	Max results
query	string	Yes	—	Natural language query
realm	string	No	—	Memory realm
skip_llm_plan	boolean	No	false	Skip LLM planning step (faster)

recall_spreadingRetrieve memories via entity graph spreading activation. Extracts capitalized words, @refs, and quoted strings from the query as entity seeds, then traverses the triplet graph (BFS depth 2, decay 0.6) to find related memories.gateway, via advanced

Parameter	Type	Required	Default	Description
limit	integer	No	10	Max results
query	string	Yes	—	Query; entity seeds extracted automatically
realm	string	No	—	Memory realm

recall_temporalSearch memories within a time window (defaults to last 7 days)via advanced

Parameter	Type	Required	Default	Description
end	string	No	—	End date
include_global	boolean	No	—	Include global memories
limit	any	No	—	Max results (default 20)
query	string	No	—	Optional semantic search query
realm	string	No	—	Filter by realm
start	string	No	—	Start date (ISO8601 or YYYY-MM-DD)

recall_true_counterfactualReturn decision points where (tool, entity) was explicitly considered and rejected. Uses the DecisionTape (Phase 10), not CDAWG sibling inference. Requires prior log_event_ex or log_decision calls.via advanced

Parameter	Type	Required	Default	Description
entity	string	Yes	—
k	any	No	—	Max results (default 5)
outcome	any	No	—	0=success 1=fail 2=error (default 0)
tool	string	Yes	—

recall_ucb1Recall with UCB1 exploration bonus — surfaces novel under-accessed memories alongside relevant onesvia advanced

Parameter	Type	Required	Default	Description
exploration	number	No	—	Exploration weight sqrt(2)≈1.414 (default)
fetch_k	any	No	—	Candidate pool size before re-ranking (default 40)
limit	any	No	—	Max results (default 10)
query	string	Yes	—	Search query
realm	string	No	—	Filter by realm

resonance_statsShow ResonanceLearner Bayesian bandit statsvia advanced

search_symbolsSemantic search for code symbolsvia advanced

Parameter	Type	Required	Default
kind	string	No	—
limit	any	No	—
project	string	No	—
query	string	Yes	—

smart_recallIntelligent memory recall with hierarchical expansion

Parameter	Type	Required	Default
expand_top	any	No	—
gwt_mode	boolean	No	—
include_global	boolean	No	—
limit	any	No	—
query	string	Yes	—
realm	string	No	—
separation_mode	boolean	No	—

stageb_set_surfaceStage B: set a memory's natural-language retrieval surface and re-embed from it (keeps telegraphic content as display). Pass surface='' to clear and re-embed from content. Returns the id and embedded flag.via advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—
surface	string	Yes	—

structured_recallThree-lens recall: facts, context, and temporal agents merged for high-fidelity retrievalvia advanced

Parameter	Type	Required	Default
limit	any	No	—
query	string	Yes	—
realm	string	No	—

Graph & Triplets

15 tools ▾

assoc_censusRead-only assoc-graph census: per-EdgeType directed-edge count plus weight histogram (<0.05 / 0.05-0.2 / 0.2-0.5 / 0.5-0.8 / >=0.8). Measure-first gate for plasticity levers.via advanced

assoc_decayGate B: decay + floor-prune one assoc EdgeType (wire numbering 0-5, default 3=CoRetrieved). Every edge weight is multiplied by factor; edges below prune_below are removed. apply=false is a dry run (counts only). One-shot saturation migration: factor=1.0, prune_below=0.2. Snapshot before apply.via advanced

Parameter	Type	Required	Default
apply	boolean	No	—
edge_type	any	No	—
factor	number	No	—
prune_below	number	No	—

connectCreate a triplet relationshipvia advanced

Parameter	Type	Required	Default	Description
object	string	Yes	—	Object entity
predicate	string	Yes	—	Relationship type
subject	string	Yes	—	Subject entity

connect_temporalCreate triplet with temporal validity

Parameter	Type	Required	Default
context_date	string	No	—
object	string	Yes	—
predicate	string	Yes	—
subject	string	Yes	—
valid_from	string	No	—
valid_to	string	No	—

cooccurrence_graphShow top co-activated memory associations for a given memoryvia advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID
limit	any	No	—	Max edges to return (default 10)

graph_pagerankPersonalized PageRank over the triplet graphvia advanced

Parameter	Type	Required	Default	Description
damping	number	No	—	Damping factor (default 0.85)
edge_types	array<string>	No	—	Predicate filter (empty = all)
iterations	any	No	—	PPR iterations (default 20)
seeds	array<string>	Yes	—	Seed nodes
top_k	any	No	—	Nodes to return (default 20)

graph_traverseBFS graph traversal from a start node over triplet edgesvia advanced

Parameter	Type	Required	Default	Description
direction	string (outgoing\|incoming\|both)	No	—	Edge direction (default outgoing)
edge_types	array<string>	No	—	Predicate filter (empty = all)
max_hops	any	No	—	Max BFS depth (default 3)
max_results	any	No	—	Max nodes returned (default 50)
start	string	Yes	—	Starting node

growAdd wisdom, belief, failure, aspiration, or dreamvia advanced

Parameter	Type	Required	Default
content	string	Yes	—
tags	string	No	—
title	string	No	—
type	string	Yes	—

queryQuery triplets with flexible filtersvia advanced

Parameter	Type	Required	Default
object	string	No	—
predicate	string	No	—
subject	string	No	—

query_graphQuery triplets by subject or object

Parameter	Type	Required	Default	Description
object	string	No	—	Query by object
subject	string	No	—	Query by subject

query_triplets_temporalQuery triplets at a point in timevia advanced

Parameter	Type	Required	Default	Description
at_date	string	No	—	YYYY-MM-DD
limit	any	No	—
object	string	No	—
predicate	string	No	—
subject	string	No	—

triplet_historyGet history of a subject-predicate relationshipvia advanced

Parameter	Type	Required	Default
limit	any	No	—
predicate	string	Yes	—
subject	string	Yes	—

triplet_query_as_ofQuery triplets for a subject valid at a given world timestamp, excluding superseded entriesvia advanced

Parameter	Type	Required	Default	Description
subject	string	Yes	—	Subject node to query
world_ms	any	No	—	World-time epoch ms (default: now)

triplet_supersedeMark one triplet as superseded by another (bi-temporal update)via advanced

Parameter	Type	Required	Default	Description
at_ms	any	No	—	Ingestion-time of supersession (default: now)
new_id	any	Yes	—	Replacing triplet ID
old_id	any	Yes	—	Triplet ID being superseded

tripletsKnowledge graph operations. Actions: connect (create temporal triplet), query (query temporal triplets), history (triplet history for a subject/object)gateway

Parameter	Type	Required	Default	Description
action	string	Yes	—	connect\|query\|history
limit	any	No	—	Max results
object	string	No	—	Object entity
predicate	string	No	—	Relationship type
subject	string	No	—	Subject entity

Code Intelligence

12 tools ▾

code_contextGet code context summary

Parameter	Type	Required	Default	Description
path	string	No	—

code_pathFind the shortest undirected connection between uniquely resolved code symbols. Use code_query IDs to disambiguate endpoints.via advanced

Parameter	Type	Required	Default
from	string	Yes	—
path	string	No	—
realm	string	No	—
to	string	Yes	—

code_queryNavigate indexed code with a natural-language question. Returns ranked symbols, file locations, typed AST/name-resolution edges and read_symbol invocations. Empty question plus path returns file context. Use before reading files.

Parameter	Type	Required	Default
limit	any	No	—
neighbors	boolean	No	—
path	string	No	—
question	string	No	—
realm	string	No	—

codebase_overviewGet full indexed codebase structure

Parameter	Type	Required	Default
format	string	No	—
include_callsites	boolean	No	—
path	string	No	—
project	string	No	—

embed_symbolsFast embed symbol metadatavia advanced

Parameter	Type	Required	Default	Description
batch_size	any	No	—
reset	boolean	No	—

Parameter	Type	Required	Default
kind	string	No	—
lang	string	No	—
limit	integer	No	—
name	string	Yes	—
path	string	No	—

learn_codebaseLearn codebase by extracting symbols. path can be a local directory or a remote git URL (https://github.com/..., [email protected]:...). Remote repos are shallow-cloned into a temp dir, indexed, then deleted.

Parameter	Type	Required	Default	Description
branch	string	No	—	Branch, tag, or commit to clone (remote only)
embed	boolean	No	—	Embed symbols synchronously (default false; use embed_symbols separately)
exclude	string	No	—
force	boolean	No	—
incremental	boolean	No	—
max_files	any	No	—
path	string	Yes	—	Local path or remote git URL
project	string	No	—	Project name (defaults to repo/dir name)

read_functionRead a function's code. Convenience wrapper for read_symbol with kind=function.gateway, via advanced

Parameter	Type	Required	Default	Description
context	integer	No	—	Lines of context before function (default: 3)
name	string	Yes	—	Function name to read
project	string	No	—	Project name filter (optional)

read_symbolRead just a symbol's code, not entire file. ~10x token savings vs full file read. Returns [kind name @ file:line-line] + code.gateway

Parameter	Type	Required	Default	Description
context	integer	No	—	Lines of context before symbol (default: 3)
kind	string	No	—	Symbol kind filter: class, function, method (optional)
line	integer	No	—	Definition line from code_query
name	string	Yes	—	Symbol name to read (e.g., 'DuckDBStore', 'daemon_call')
path	string	No	—	Exact path from code_query
project	string	No	—	Project name filter (optional)

symbol_calleesFind all symbols that a symbol calls. Queries triplets where predicate=calls and subject=symbol.gateway, via advanced

Parameter	Type	Required	Default	Description
limit	integer	No	—	Max results (default: 20)
name	string	Yes	—	Symbol name to find callees for

symbol_callersFind all callers of a symbol without grep. Queries triplets where predicate=calls and object=symbol.gateway, via advanced

Parameter	Type	Required	Default	Description
limit	integer	No	—	Max results (default: 20)
name	string	Yes	—	Symbol name to find callers for

symbol_event_logQuery the symbol-keyed event log. Filter by symbol_name and/or file_path.via advanced

Parameter	Type	Required	Default	Description
file_path	string	No	—	Filter by file path
limit	any	No	—	Max events to return (default 50)
symbol_name	string	No	—	Filter by symbol name

Context & Status

16 tools ▾

askNatural language insight query: retrieves and synthesizes memories to answer a question about the user, session, or projectvia advanced

Parameter	Type	Required	Default
limit	any	No	—
question	string	Yes	—
realm	string	No	—

checkpointSave session state

Parameter	Type	Required	Default
active_files	array<string>	No	—
discoveries	array<string>	No	—
mood	string	No	—
next_steps	array<string>	No	—
realm	string	No	—
summary	string	No	—

compact_contextMemory-aware context compaction. Scores conversation messages by recency, semantic relevance to query, and memory coverage (content already in memory is safer to drop). Returns a subset of messages fitting the target token budget.via advanced

Parameter	Type	Required	Default	Description
distill_novel	boolean	No	—	Reserved for future use
messages	array<object>	Yes	—	Conversation messages [{role,content}]
query	string	No	—	Upcoming task hint for semantic scoring
target_ratio	number	No	—	Fraction of tokens to KEEP (default 0.4)

explore_expandGet full content of a memory

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID

explore_neighborsGet nodes connected via triplets

Parameter	Type	Required	Default	Description
direction	string	No	—	outgoing, incoming, or both
node	string	Yes	—	Node name

explore_peekGet summary of a memory (first 200 chars)

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID

explore_recallLightweight recall - titles/scores only

Parameter	Type	Required	Default	Description
limit	any	No	—	Max results (default 10)
query	string	Yes	—	Search query

get_policiesGet active policiesvia advanced

Parameter	Type	Required	Default
limit	any	No	—
scope	string	No	—
type	string	No	—

impl_startStart self-improvement implementation sadhanavia advanced

Parameter	Type	Required	Default
interval_seconds	any	No	—
max_turns	any	No	—
realm	string	No	—
repo	string	No	—

list_policiesList CEC intervention policies (shadow and active). Each entry shows rule source, kind (OpenTask/TurnInjection/GuardPolicy), shadow event count, lift, and fire count.via advanced

Parameter	Type	Required	Default	Description
active_only	boolean	No	—	If true, only return active (promoted) policies

observeStore an observation/learning (SSL v0.4)

Parameter	Type	Required	Default	Description
arousal	number	No	—	Affect arousal: 0.0 to 1.0
category	string	No	—
confidence	number	No	—
content	string	Yes	—
derivation	string	No	—	SSL v0.4 <=@ provenance: comma-separated source memory IDs this was abstracted from (required at G:1+)
flags	string	No	—	Structural flags: ORIGIN,CORE,PIVOT,GENESIS,TURNING
granularity	any	No	—	SSL v0.4 granularity tier: 0=atom,1=episode,2=claim,3=operator,4=boundary
refs	string	No	—	Cross-references: comma-separated tag names or memory IDs
source_loc	string	No	—	SSL v0.4 src: external source grounding, e.g. file:line or doc section
tags	string	No	—
title	string	Yes	—
valence	number	No	—	Affect valence: -1.0 to +1.0

smart_contextBuild intelligent context. Modes: fast (<80ms), full (<200ms), rlm (RLM-style dynamic exploration via soul_repl). With resolver_mode=true (default), prepends digest-node and decision memories before code symbols.gateway

Parameter	Type	Required	Default	Description
code	boolean	No	—	Include code symbols (default: true)
limit	integer	No	—	Token limit (default: 300)
memories	boolean	No	—	Include semantic memories (default: true)
mode	string (fast\|full\|rlm)	No	—	fast: <80ms (vector + BM25), full: <200ms (full_resonate), rlm: dynamic exploration via soul_repl
neighbors	boolean	No	—	Include triplet neighbors (default: true)
realm	string	No	—	Filter by realm
resolver_mode	boolean	No	—	Prepend digest-node and decision memories before code symbols (default: true)
task	string	Yes	—	Query to find context for

soul_contextGet current soul state and statistics

soul_replRLM-style Python REPL for programmatic memory exploration. Write code with soul.* methods: search(), recall(), expand(), triplets(), recent(), remember(), symbols(). Supports persistent sessions via session_id. Call with no code for API reference.gateway, via advanced

Parameter	Type	Required	Default	Description
code	string	No	—	Python code to execute. Has access to soul.search(), soul.recall(), soul.expand(), etc.
reset	boolean	No	—	Reset namespace before execution (default: false)
session_id	string	No	—	Session name for persistent state — variables survive across calls with the same session_id

think_wanderTrigger internal memory synthesisvia advanced

Parameter	Type	Required	Default	Description
realm	string	No	—

trajectory_compactAttention-weighted turn selection from a transcript. Embeds each turn, scores by cosine similarity to the task description, applies MAD adaptive threshold, enforces token budget. Returns a lossless subset of the most task-relevant turns.via advanced

Parameter	Type	Required	Default	Description
budget_tokens	any	No	—	Target token budget (default 4000)
include_system	boolean	No	—	Include system turns (default false)
mad_k	number	No	—	MAD threshold multiplier (default 1.5, lower=more turns)
path	string	No	—	Direct path to JSONL transcript
role_filter	string	No	—	Filter by role: user, assistant, or empty for all
session_id	string	No	—	Session ID (auto-finds transcript)
task	string	Yes	—	What the downstream agent needs to accomplish

Realms

9 tools ▾

realm_addAdd memory to a shared realm (stub)via advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—
realm	string	Yes	—

realm_detectDetect current realm from environment

realm_getGet all realms a memory belongs tovia advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—

realm_listList all known realmsvia advanced

realm_removeRemove memory from a shared realm (stub)via advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—
realm	string	Yes	—

realm_setSet primary realm for a memoryvia advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—
realm	string	Yes	—

realm_visibilitySet visibility level (stub)via advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—
visibility	any	Yes	—

remap_realmsBulk-remap realms per {old:new} mapping (mapping_file=path or mapping=object; dry_run defaults true)via advanced

Parameter	Type	Required	Default
dry_run	boolean	No	—
mapping	object	No	—
mapping_file	string	No	—

trim_realm_namesFix realm names with trailing whitespace/newlinesvia advanced

Sessions & Continuity

19 tools ▾

get_turnsGet conversation turns for a sessionvia advanced

Parameter	Type	Required	Default
limit	any	No	—
session_id	string	No	—
start_index	any	No	—

ledger_deleteDelete checkpointvia advanced

Parameter	Type	Required	Default	Description
key	string	No	—	Canonical checkpoint key (session:project)
project	string	No	—
session_id	string	No	—

ledger_getGet checkpoint by key or session/project

Parameter	Type	Required	Default	Description
include_snapshot	boolean	No	—	Include snapshot preview (truncated) in response
key	string	No	—	Canonical checkpoint key (session:project)
project	string	No	—
session_id	string	No	—

ledger_healthGet ledger event counts by kind and queue health metricsvia advanced

ledger_listList recent checkpoints

Parameter	Type	Required	Default	Description
limit	any	No	—
project	string	No	—

ledger_loadLoad most recent checkpoint

Parameter	Type	Required	Default	Description
include_snapshot	boolean	No	—	Include snapshot preview (truncated) in response
project	string	No	—
session_id	string	No	—

ledger_saveSave session checkpoint

Parameter	Type	Required	Default
active_files	array<string>	No	—
blockers	array<string>	No	—
coherence	number	No	—
confidence	number	No	—
decisions	array<string>	No	—
discoveries	array<string>	No	—
mood	string	No	—
next_steps	array<string>	No	—
project	string	No	—
session_id	string	No	—
snapshot	string	No	—
todos	array	No	—
transcript_path	string	No	—

long_task_activeGet active long-running task

Parameter	Type	Required	Default	Description
realm	string	No	—

long_task_completeMark task as completed

Parameter	Type	Required	Default	Description
outcome	string	Yes	—
task_id	string	Yes	—

long_task_evaluateEvaluate task completionvia advanced

Parameter	Type	Required	Default	Description
task_id	string	Yes	—

long_task_eventAppend event to task log

Parameter	Type	Required	Default
kind	string	Yes	—
payload	string	No	—
related_entities	array<string>	No	—
tags	array<string>	No	—
task_id	string	Yes	—

long_task_getGet a long-running task by IDvia advanced

Parameter	Type	Required	Default	Description
task_id	string	Yes	—

long_task_snapshotGet synthesized task context

Parameter	Type	Required	Default
max_tokens	any	No	—
mode	string	No	—
task_id	string	Yes	—

long_task_startStart a long-running task

Parameter	Type	Required	Default
goal	string	Yes	—
hard_checks	array<string>	No	—
realm	string	No	—
soft_checks	array<string>	No	—
task_id	string	Yes	—
work_items	array<string>	No	—

long_task_updateUpdate long-running task progress

Parameter	Type	Required	Default
blockers	array<string>	No	—
completed_summary	string	No	—
task_id	string	Yes	—
work_items	array<string>	No	—

read_transcriptRead JSONL transcript with paginationvia advanced

Parameter	Type	Required	Default
keyword	string	No	—
limit	any	No	—
max_chars_per_turn	any	No	—
metadata_only	boolean	No	—
path	string	No	—
role_filter	string	No	—
session_id	string	No	—
start_turn	any	No	—

session_listList active sessionsvia advanced

Parameter	Type	Required	Default	Description
realm	string	No	—
status	string	No	—

session_syncSync session registryvia advanced

Parameter	Type	Required	Default	Description
projects_dir	string	No	—

task_stateDeterministic task-state check (capability #3): what is the current state of task X? Exact keyed lookup of the LATEST stored [task] record by its slug — bypasses fuzzy recall so an agent resuming a discontinuous session gets the durable status/next-step. Returns found + the record.via advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—	Task slug (the id: token of the [task] record)

Cross-Harness Messaging

6 tools ▾

cross_harness_conflictsFind memories where claude-code and codex harnesses disagree on the same topicvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max conflict pairs to return (default 20)
min_disagreement_score	number	No	—	Minimum 1-cosine_similarity threshold (default 0.3)
realm	string	No	—	Realm filter (default: all)

msg_ack_allAcknowledge all messagesvia advanced

Parameter	Type	Required	Default	Description
session_id	string	No	—

msg_historyGet message historyvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—
session_id	string	No	—

msg_inboxCheck unread messagesvia advanced

Parameter	Type	Required	Default
auto_ack	boolean	No	—
limit	any	No	—
min_priority	any	No	—
session_id	string	No	—

msg_respondReply to a message using the original sender/target from the eventvia advanced

Parameter	Type	Required	Default	Description
content	string	Yes	—	Reply content
message_id	any	Yes	—	Event ID of the message to reply to
session_id	string	No	—	Override sender session_id (defaults to original target)

msg_sendSend message to another session

Parameter	Type	Required	Default
content	string	Yes	—
content_type	string	No	—
priority	any	No	—
session_id	string	No	—
target	string	Yes	—
target_type	string	No	—
ttl	any	No	—

Narrative & Work Modes

5 tools ▾

log_decisionLog a decision point to the DecisionTape: chosen action + alternatives considered and rejected. Enables recall_true_counterfactual.via advanced

Parameter	Type	Required	Default	Description
chosen_entity	string	Yes	—
chosen_outcome	any	No	—	0=success 1=fail 2=error 3=partial
chosen_tool	string	Yes	—
confidence_delta	number	No	—	chosen_confidence - best_alternative_confidence
rejected_json	string	No	—	JSON array of [sym_u64, rejection_reason_u8] pairs
ts_ms	any	No	—

log_exposureLog memory exposure (SUS)via advanced

Parameter	Type	Required	Default
hook_type	string	Yes	—
memory_ids	array<any>	Yes	—
ranks	array<any>	No	—
resonance_scores	array<number>	No	—
session_id	string	Yes	—
turn_id	any	Yes	—

narrative_historyGet work mode segment historyvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—
session_id	string	No	—

narrative_logAppend event to session logvia advanced

Parameter	Type	Required	Default
files_mentioned	string	No	—
kind	string	Yes	—
payload	string	No	—
session_id	string	No	—
success	boolean	No	—
summary	string	Yes	—
tool_name	string	No	—

narrative_statusGet work mode and segment summaryvia advanced

Parameter	Type	Required	Default	Description
session_id	string	No	—

Distillation & Embeddings

13 tools ▾

compact_walCompact WAL: save full snapshot then delete covered segmentsvia advanced

densify_backfill#13 retro-backfill of SameSession edges over existing session-tagged memories (same K=3 decaying-weight rule as the write-time hook). apply=false is a dry run: reports sessions, memories, sibling pairs, and a group-size histogram, writing nothing. apply=true creates the edges (idempotent; rollback via remove_assoc_edges_by_type).via advanced

Parameter	Type	Required	Default	Description
apply	boolean	No	—

distill_nowSynchronously distill ONE session now and return the counts (learnings + value-facts stored/deduped). Runs the lock-fixed distill path in-daemon; recall stays responsive.via advanced

Parameter	Type	Required	Default	Description
realm	string	No	—	Optional realm (default brahman or the registered realm)
session_id	string	Yes	—	Session ID to distill
transcript_path	string	No	—	Optional JSONL path; registers it if given, else resolves from durable transcript/register

distill_set_modelChange distillation modelvia advanced

Parameter	Type	Required	Default	Description
enabled	boolean	No	—
model	string	Yes	—

embed_coverageSemantic-index coverage: how many memories SHOULD have a vector vs how many DO. pending_count cannot answer this — the embed queue drains on failure as well as success, so a lost embedding leaves the queue empty.via advanced

embed_probeFor a memory id: cosine of its STORED vector against a fresh single-prefix embed vs a double-prefix embed of the same content. ~1.0 identifies which vector space the memory actually lives in.via advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—

flush_embeddingsFlush the pending embedding queue synchronously — call after remember_batch to ensure HNSW/semantic recall is available immediately. Returns {flushed: N}.via advanced

get_embeddingsBatch-fetch raw embedding vectors for memory IDs. Returns JSON object mapping id→vector. Used for computing S-entropy (Gram matrix effective rank) over recall candidates.via advanced

Parameter	Type	Required	Default	Description
ids	array<string>	Yes	—	Array of memory ID strings

health_checkCheck daemon health

pending_embed_idsReturn IDs of memories awaiting embedding (stuck embed queue)via advanced

queue_statusLive work-queue depth (pending in-flight + unclaimed), processed/distilled/failed counters, pending embeddings, dead-lettersvia advanced

rebuild_fts_indexRebuild FTS index for BM25 searchvia advanced

semantic_backfillDense-kNN SemanticNeighbor edge backfill: for each non-deleted memory, link its top-k realm-scoped HNSW neighbors (cosine>=min_cos) with bidirectional similarity edges (wire 6). The first genuine memory<->memory knowledge relation in the assoc graph, distinct from CoRetrieved's retrieval-history prior. apply=false is a dry run (counts candidate edges, no writes). apply=true creates them (idempotent; rollback via remove_assoc_edges_by_type edge_type=6).via advanced

Parameter	Type	Required	Default
apply	boolean	No	—
k	any	No	—
min_cos	number	No	—

Consolidation & Contradictions

15 tools ▾

conflict_inspectorSemantic search + show status and contradiction partners for each hitvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max memories to scan (default 10)
query	string	Yes	—	Search query
realm	string	No	—	Filter by realm

consolidate_similarMerge near-duplicate memories — keeps stronger, soft-deletes weakervia advanced

Parameter	Type	Required	Default	Description
dry_run	boolean	No	—	Preview without deleting (default true)
limit	any	No	—	Max pairs to merge (default 10)
realm	string	No	—	Filter by realm
threshold	number	No	—	Similarity threshold (default 0.92)

consolidation_passRun Sequitur grammar consolidation: find frequent bigrams in EventTape and promote rules to the triplet KG (subject=rule:..., predicates: compresses/avg_outcome/support/tape_range).via advanced

detect_contradictionsDetect contradictions for a stored memory against realm peersvia advanced

Parameter	Type	Required	Default	Description
memory_id	string	Yes	—	Memory ID to check
realm	string	No	—	Realm to scan (default global)

disable_sourceAdd a source to the deny-list via tripletvia advanced

Parameter	Type	Required	Default	Description
source	string	Yes	—	Source identifier to deny

find_near_duplicatesFind memory pairs with high semantic similarity (near-duplicates)via advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max pairs to return (default 20)
realm	string	No	—	Filter by realm
threshold	number	No	—	Cosine similarity threshold (default 0.90)

hygiene_statsGet memory hygiene statisticsvia advanced

reconsolidateUpdate content of a memory during its labile window (reconsolidation)via advanced

Parameter	Type	Required	Default	Description
content	string	Yes	—	New/corrected content
id	string	Yes	—	Memory ID to update
reason	string	No	—	Optional reason for reconsolidation

resolve_contradictionResolve a contradiction: declare winner supersedes loser, store CORRECTION memoryvia advanced

Parameter	Type	Required	Default	Description
loser_id	string	Yes	—	Memory ID to demote
reason	string	No	—	Explanation for the resolution
winner_id	string	Yes	—	Memory ID that is correct

save_spectral_snapshotSave spectral stats snapshot for drift trackingvia advanced

scan_contradictionsBackground scan: find contradiction candidates across all memories in a realmvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max candidates to return (default 50)
realm	string	No	—	Realm to scan

show_conflictsSemantic search + show contradiction pairs for matching memoriesvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max memories to scan (default 20)
query	string	Yes	—	Search query
realm	string	No	—	Filter by realm

spectral_driftCompare current embedding geometry with last snapshotvia advanced

what_supersededShow the full supersession chain for a memoryvia advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID to trace

why_activeExplain why a memory is active: status, epistemic source, confirmations, contradictionsvia advanced

Parameter	Type	Required	Default	Description
id	string	Yes	—	Memory ID to inspect

Themes

4 tools ▾

theme_getGet theme detailsvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—

theme_listList all themes with statisticsvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—
realm	string	No	—

theme_recallTwo-stage theme-based retrievalvia advanced

Parameter	Type	Required	Default
limit	any	No	—
query	string	Yes	—
realm	string	No	—

theme_statsGet theme organization statisticsvia advanced

Parameter	Type	Required	Default	Description
realm	string	No	—

Provenance & Verification

13 tools ▾

behavioral_probeScore text against behavioral centroid clusters. Returns per-class similarity scores (sycophantic/hedging/shallow/direct) and overall quality estimate. Requires prior probe_seed calls.via advanced

Parameter	Type	Required	Default	Description
text	string	Yes	—	Text to probe (e.g. a Claude response)

calibration_recordRecord prediction outcomevia advanced

Parameter	Type	Required	Default	Description
domain	string	Yes	—
success	boolean	Yes	—

calibration_scoreGet accuracy scorevia advanced

Parameter	Type	Required	Default	Description
domain	string	No	—

correction_checkDeterministic durable-correction check (capability #2): does any stored [correction] trigger recur in this turn? Exact keyed bigram probe of the turn text against corrected-mistake phrases — bypasses fuzzy recall (which loses corrections on cosine similarity ~99% of the time). Returns found + the correction(s), newest first (latest-wins).

Parameter	Type	Required	Default	Description
text	string	Yes	—	Turn / context text to scan for a recurring corrected mistake

get_relationship_eventsGet relationship eventsvia advanced

Parameter	Type	Required	Default
event_type	string	No	—
limit	any	No	—
session_id	string	No	—

probe_calibrateAdd a confirmed exemplar to a behavioral class to refine its centroid. Use when you have a clear example of the behavior.via advanced

Parameter	Type	Required	Default	Description
class	string	Yes	—	Behavioral class to update
text	string	Yes	—	Confirmed exemplar text

probe_seedStore an exemplar text as a centroid for a behavioral class (sycophantic/hedging/shallow/direct). Bootstrap the probe with representative examples.via advanced

Parameter	Type	Required	Default	Description
class	string	Yes	—	Behavioral class: sycophantic \| hedging \| shallow \| direct
note	string	No	—	Optional annotation
text	string	Yes	—	Exemplar text for this class

probe_statusShow how many exemplars exist per behavioral class. Use to verify the probe is seeded before running behavioral_probe.via advanced

provenance_checkDeterministic anti-reprocessing check: has this file/task already been processed? Exact keyed lookup of a prior [done] record by content-hash and/or input path — bypasses fuzzy recall. Returns found + the record.via advanced

Parameter	Type	Required	Default	Description
input	string	No	—	Input path (fallback handle)
sha	string	No	—	Content hash of the input (tried first — content identity)

span_backfillBackfill the span lane over all transcripts under projects_dir (default ~/.claude/projects). Incremental + idempotent via per-file watermarks. Reports unique atoms, new atoms, redaction count.via advanced

Parameter	Type	Required	Default	Description
projects_dir	string	No	—	Transcript root (default ~/.claude/projects)

span_backfill_memoriesBackfill the memory↔span edge: run the atom extractor over every live memory's text so distilled beliefs link to the verbatim paths/commands/ids they mention. Idempotent (per-memory content hash); superseded memories relink. Reports memories linked + new spans.via advanced

span_queryRetrieve verbatim transcript atoms (file paths, URLs, file:line locators, bash commands, error signatures) captured across all sessions. No LLM/GPU: exact-substring recall over a deduplicated span index. Realm-scoped by default to prevent cross-project bleed. Use when you need an exact locator you saw before, not a paraphrase.via advanced

Parameter	Type	Required	Default	Description
k	any	No	—	Max atoms to return (default 6)
query	string	Yes	—	Substring or token to match against stored atoms
realm	string	No	—	Restrict to this realm (project); empty = unscoped

span_statsReport span lane size: unique atoms, on-disk bytes, total redactions.via advanced

Sadhana

12 tools ▾

sadhanaAutonomous background learning loop control. Actions: start, stop, pause, resume, status, list, checkpoint, set_goal, set_interval, set_modelgateway

Parameter	Type	Required	Default	Description
action	string	Yes	—	start\|stop\|pause\|resume\|status\|list\|checkpoint\|set_goal\|set_interval\|set_model
goal	string	No	—	Goal text (for set_goal)
id	any	No	—	Sadhana ID (for stop/pause/resume/status/checkpoint)
interval	any	No	—	Interval in seconds (for set_interval)
max_turns	any	No	—	Max turns (for start)
model	string	No	—	Model name (for set_model/start)

sadhana_checkpointReport mid-cycle checkpointvia advanced

Parameter	Type	Required	Default
id	any	Yes	—
status	string (progressed\|achieved\|blocked)	Yes	—
summary	string	Yes	—

sadhana_listList sadhanasvia advanced

Parameter	Type	Required	Default
limit	any	No	—
realm	string	No	—
state	string	No	—

sadhana_pausePause a sadhanavia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—

sadhana_resumeResume a paused sadhanavia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—

sadhana_set_goalChange sadhana goalvia advanced

Parameter	Type	Required	Default	Description
goal	string	Yes	—
id	any	Yes	—

sadhana_set_intervalChange tick intervalvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—
interval	any	Yes	—

sadhana_set_max_turnsSet max turns per cyclevia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—
max_turns	any	Yes	—

sadhana_set_modelChange brain modelvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—
model	string	Yes	—

sadhana_startCreate and start an autonomous agentvia advanced

Parameter	Type	Required	Default
brain_model	string	No	—
brain_provider	string	No	—
goal	string	Yes	—
goal_dsl	object	No	—
interval_seconds	any	No	—
max_turns	any	No	—
realm	string	No	—

sadhana_statusGet sadhana statusvia advanced

Parameter	Type	Required	Default	Description
history_limit	any	No	—
id	any	Yes	—

sadhana_stopStop a sadhanavia advanced

Parameter	Type	Required	Default
id	any	Yes	—
reason	string	No	—
success	boolean	No	—

Dreams & Curiosity

13 tools ▾

curiosity_gapsList knowledge gapsvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—
realm	string	No	—

curiosity_note_gapRecord a knowledge gapvia advanced

Parameter	Type	Required	Default
context	string	No	—
gap	string	Yes	—
realm	string	No	—

curiosity_resolveMark gap as resolvedvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—
learned	string	No	—

dream_cancelCancel a dreamvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—

dream_force_wokeForce stuck dream to wokevia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—

dream_listList recent dreams

Parameter	Type	Required	Default	Description
limit	any	No	—
realm	string	No	—

dream_startStart an autonomous dream

Parameter	Type	Required	Default	Description
brain_model	string	No	—	Model name, e.g. gemma4:26b or sonnet
brain_provider	string	No	—	Brain provider: claude or local
publish_path	string	No	—
realm	string	No	—
topic	string	Yes	—

dream_statusGet dream details

Parameter	Type	Required	Default	Description
id	any	Yes	—

dream_wanderAuto-select topic and dream

Parameter	Type	Required	Default	Description
publish_path	string	No	—
realm	string	No	—

researchCuriosity-driven research. Actions: cycle (get topic to research), topics (list research topics), store (save findings)gateway

Parameter	Type	Required	Default	Description
action	string	Yes	—	cycle\|topics\|store
findings	string	No	—	Findings (store)
gap_id	any	No	—	Gap ID to resolve (store)
limit	any	No	—	Max results
realm	string	No	—	Filter by realm
source	string	No	—	Topic source: gaps\|weak\|suggest (topics)
sources	array<string>	No	—	URLs (store)
topic	string	No	—	Topic (store)

research_cycleRun one curiosity-driven research cycle. Returns a topic to research with context. After calling this, use WebSearch to find information, then call research_store with results.gateway, via advanced

Parameter	Type	Required	Default	Description
realm	string	No	—	Filter by realm/project

research_storeStore research results as memories with source attribution. Call after using WebSearch to learn about a topic.gateway, via advanced

Parameter	Type	Required	Default	Description
confidence	number	No	—	Confidence in findings 0-1 (default: 0.7)
findings	string	Yes	—	Key learnings in SSL format
gap_id	integer	No	—	Curiosity gap ID to resolve (optional)
sources	array<string>	No	—	URLs or references
topic	string	Yes	—	What was researched

research_topicsGet topics that need research. Returns curiosity gaps, low-confidence memories, or suggested topics. Use WebSearch to research these, then store results with research_store.gateway, via advanced

Parameter	Type	Required	Default	Description
limit	integer	No	—	Max topics to return (default: 3)
realm	string	No	—	Filter by realm/project
source	string	No	—	Where to get topics: gaps (curiosity gaps), weak (low-confidence memories), suggest (AI-suggested based on recent work)

Learning & Outcomes

18 tools ▾

effective_scorer_weightsShow effective scoring weights — baseline + learned deltas for all factorsvia advanced

get_source_weightsView learned recall source weights — how much each source is trusted per domainvia advanced

Parameter	Type	Required	Default	Description
domain	string	No	—	Filter by domain (omit for all)

get_sus_metricsGet Soul Utility Score metricsvia advanced

Parameter	Type	Required	Default	Description
days	any	No	—

integration_statsPer-source success rates and learned weights across all domainsvia advanced

learnStore learning in soul memory. Types: correction (wrong→right fix), preference (user style), insight (generalizable wisdom), approach (what worked in a state), outcome (did suggestion help?), milestone (achievement), analysis (reproducible analysis with data/scripts)gateway

Parameter	Type	Required	Default	Description
approach	string	No	—	Approach text (approach)
category	string	No	—	Category (preference)
context	string	No	—	Context for the learning
correct	string	No	—	What is correct (correction)
data_paths	array<string>	No	—	Data paths (analysis)
description	string	No	—	Description (analysis)
details	string	No	—	Details (outcome)
domain	string	No	—	Domain (insight/analysis)
findings	string	No	—	Findings (analysis)
helped	boolean	No	—	Did it help? (outcome)
insight	string	No	—	Insight text (insight)
milestone	string	No	—	Milestone text (milestone)
name	string	No	—	Analysis name (analysis)
outcome	string	No	—	Outcome (approach)
preference	string	No	—	Preference text (preference)
project	string	No	—	Project (analysis)
script_paths	array<string>	No	—	Script paths (analysis)
state	string	No	—	State/mood (approach)
suggestion	string	No	—	Suggestion (outcome)
type	string	Yes	—	correction\|preference\|insight\|approach\|outcome\|milestone\|analysis
wrong	string	No	—	What was wrong (correction)

learn_analysisRecord an analysis with data and script locations. Makes analyses reproducible and findable later. Use after completing any significant analysis.gateway, via advanced

Parameter	Type	Required	Default	Description
data_paths	array<string>	No	—	Paths to input data files/directories
description	string	No	—	What the analysis does/investigates
findings	string	No	—	Key findings or results (optional)
name	string	Yes	—	Short name for the analysis (e.g., 'AGP codon optimization', 'bin_28 damage patterns')
project	string	No	—	Project name for organization (optional)
script_paths	array<string>	No	—	Paths to analysis scripts

learn_approachStore what approach worked when in a particular state/mood. Builds emotional memory for adapting to session dynamics.gateway, via advanced

Parameter	Type	Required	Default	Description
approach	string	Yes	—	What helped in this state (e.g., 'step back and reread requirements')
outcome	string	No	—	What happened after (optional)
state	string	Yes	—	State/mood: stuck, debugging, exploring, flowing, frustrated, uncertain, rushing

learn_correctionStore a correction when I was wrong. Creates high-confidence counter-memory with 'corrects' triplet linking to original mistake.gateway, via advanced

Parameter	Type	Required	Default	Description
context	string	No	—	Context where this applies (optional)
correct	string	Yes	—	The correct information/approach
wrong	string	Yes	—	What I said/did that was incorrect

learn_insightStore a generalizable insight that applies across projects. Use for patterns, techniques, and wisdom not tied to specific codebase.gateway, via advanced

Parameter	Type	Required	Default	Description
domain	string	Yes	—	Domain: programming, debugging, architecture, testing, performance, security, communication
insight	string	Yes	—	The generalizable insight or pattern
learned_from	string	No	—	Context where this was learned (optional, for future reference)

learn_milestoneRecord a relationship milestone - achievements, personal context, significant moments worth remembering.gateway, via advanced

Parameter	Type	Required	Default	Description
date	string	No	—	When it happened (optional, defaults to now)
milestone	string	Yes	—	What happened (e.g., 'shipped v1.0', 'first successful release')
significance	string	No	—	Why it matters (optional)

learn_outcomeRecord whether a suggestion/approach actually helped. Builds feedback loop for improving future suggestions.gateway, via advanced

Parameter	Type	Required	Default	Description
details	string	No	—	Why it helped or didn't (optional but valuable)
helped	boolean	Yes	—	Did it help? true/false
suggestion	string	Yes	—	What was suggested or tried

learn_preferenceStore a user preference for adapting communication/behavior. Global visibility so it applies across all projects.gateway, via advanced

Parameter	Type	Required	Default	Description
category	string	Yes	—	Preference category: communication, detail, autonomy, style, workflow
example	string	No	—	Example demonstrating this preference (optional)
preference	string	Yes	—	The preference to remember (e.g., 'prefers concise responses')

learned_scorer_statsCurrent learned scoring model — version, factor count, loss, outcome countvia advanced

metacognition_evaluateSelf-evaluate learning effectivenessvia advanced

record_feedbackRecord whether a recall source was useful — updates learned source weightsvia advanced

Parameter	Type	Required	Default	Description
query_domain	string	No	—	Domain of the query (default general)
source	string	Yes	—	Source: semantic, keyword, temporal, artifact, association (required)
was_useful	boolean	No	—	Whether the source's results were useful (default true)

surprise_learning_statsRolling surprise credit stats — tracked memories, gates passed, strength adjustmentsvia advanced

update_scorer_modelApply learned weight deltas to the scoring model from outcome calibrationvia advanced

Parameter	Type	Required	Default	Description
mean_loss	number	No	—	EWMA loss from calibration
model_version	any	No	—	Monotonic version number
outcome_count	any	No	—	Total outcomes used for calibration
weights	object	No	—	Factor name → {delta, min_delta, max_delta} learned adjustments

update_source_weightManually override a recall source weightvia advanced

Parameter	Type	Required	Default	Description
domain	string	No	—	Domain (default general)
source	string	Yes	—	Source name (required)
weight	number	No	—	New weight [0-2] (default 1.0)

Anticipation & Habits

15 tools ▾

anticipation_filterGet predictions passing annoyance gatevia advanced

Parameter	Type	Required	Default	Description
max	any	No	—
session_id	string	No	—

anticipation_listList learned anticipation patternsvia advanced

Parameter	Type	Required	Default
limit	any	No	—
realm	string	No	—
sort_by	string	No	—

anticipation_observeRecord context->action patternvia advanced

Parameter	Type	Required	Default
action	string	Yes	—
context	string	Yes	—
realm	string	No	—

anticipation_predictPredict likely actionsvia advanced

Parameter	Type	Required	Default
context	string	Yes	—
limit	any	No	—
min_confidence	number	No	—
realm	string	No	—

anticipation_successMark prediction as successfulvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—

habit_listList formed habits

Parameter	Type	Required	Default
limit	any	No	—
min_strength	number	No	—
realm	string	No	—
sort_by	string	No	—

habit_matchFind matching habits

Parameter	Type	Required	Default
context	string	Yes	—
min_strength	number	No	—
realm	string	No	—

habit_observeRecord trigger->response patternvia advanced

Parameter	Type	Required	Default
realm	string	No	—
response	string	Yes	—
trigger	string	Yes	—

habit_strengthenStrengthen a habit

Parameter	Type	Required	Default	Description
amount	number	No	—
id	any	Yes	—

habit_weakenWeaken a habitvia advanced

Parameter	Type	Required	Default	Description
amount	number	No	—
id	any	Yes	—

predict_neededGet predicted next-needed memories from the Markov chain access predictorvia advanced

Parameter	Type	Required	Default	Description
k	any	No	—	Number of predictions (default 8)

trigger_addCreate a trigger automaton (prospective memory). Arms on creation, fires when conditions met or tension exceeds threshold.via advanced

Parameter	Type	Required	Default	Description
action	object	Yes	—	Action on fire (Notify, InjectMemory, EmitEvent, RememberFact)
condition	object	Yes	—	Trigger condition (TimeAfter, ConstraintMatch, EventMatch, AllOf, AnyOf)
deadline_ms	any	No	—	Deadline timestamp ms (0=no deadline)
gain	number	No	—	Emotional importance 0-1 (default 0.5)
name	string	Yes	—	Human-readable trigger name
realm	string	No	—	Realm scope (default: global)
tension_threshold	number	No	—	Tension level to auto-fire (default 0.8)

trigger_dismissExpire/dismiss a trigger without firingvia advanced

Parameter	Type	Required	Default	Description
trigger_id	any	Yes	—	Trigger ID to dismiss

trigger_fireManually fire a triggervia advanced

Parameter	Type	Required	Default	Description
trigger_id	any	Yes	—	Trigger ID to fire

trigger_listList all triggers with their statusvia advanced

Profile & Goals

8 tools ▾

goal_completeMark goal completedvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—
outcome	string	Yes	—

goal_getGet goal by IDvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—

goal_listList goalsvia advanced

Parameter	Type	Required	Default
limit	any	No	—
realm	string	No	—
sort_by	string	No	—
status	string	No	—

goal_progressUpdate goal progressvia advanced

Parameter	Type	Required	Default
id	any	Yes	—
milestone	string	No	—
progress	number	Yes	—

goal_setDefine a long-term goalvia advanced

Parameter	Type	Required	Default
deadline	any	No	—
description	string	No	—
milestones	string	No	—
realm	string	No	—
title	string	Yes	—

profile_getGet user profilevia advanced

Parameter	Type	Required	Default	Description
user_id	string	No	—

profile_observeRecord user observationvia advanced

Parameter	Type	Required	Default
observation_type	string	Yes	—
user_id	string	No	—
value	string	Yes	—

profile_updateUpdate profile fieldvia advanced

Parameter	Type	Required	Default
field	string	Yes	—
user_id	string	No	—
value	string	Yes	—

Skill & Agent Registry

19 tools ▾

add_delegationRecord a delegation edge — tracks which agent handed off to which, with optional handoff notevia advanced

Parameter	Type	Required	Default	Description
from_agent	string	Yes	—	Delegating agent name
handoff_note	string	No	—	Optional context passed at handoff
task_id	any	Yes	—	Task ID
to_agent	string	Yes	—	Receiving agent name

add_probeAdd a pending probe — an open question that must be answered to unblock or complete a taskvia advanced

Parameter	Type	Required	Default	Description
expected_answerer	string	No	—	Agent or role expected to answer
priority	any	No	—	Priority 1-10 (default 5)
question	string	Yes	—	Open question to be answered
task_id	any	Yes	—	Task ID

agent_disableDisable (revoke) an agent.gateway, via advanced

Parameter	Type	Required	Default	Description
agent_id	string	Yes	—	Agent identifier to disable

agent_getGet an agent's identity record.gateway, via advanced

Parameter	Type	Required	Default	Description
agent_id	string	Yes	—	Agent identifier

agent_listList all registered agents.gateway, via advanced

agent_protocol_statsShow agent protocol memory statistics — total tasks, delegations, evidence links, probes, criteria countsvia advanced

agent_upsertRegister or update an agent identity in the multi-agent registry.gateway, via advanced

Parameter	Type	Required	Default	Description
agent_id	string	Yes	—	Unique agent identifier
description	string	No	—	What this agent does
display_name	string	No	—	Human-readable name

get_taskGet full task view — contract, delegations, evidence links, pending probes, and completion criteriavia advanced

Parameter	Type	Required	Default	Description
task_id	any	Yes	—	Task ID

link_evidenceLink a memory to a task as evidence — records which agent produced it and evidence kindvia advanced

Parameter	Type	Required	Default	Description
evidence_kind	any	No	—	0=Observation 1=Artifact 2=Result 3=Analysis 4=UserFeedback
memory_id	any	Yes	—	Memory ID to link
produced_by	string	No	—	Agent that produced this evidence
relevance	number	No	—	Relevance score 0-1 (default 1.0)
task_id	any	Yes	—	Task ID

query_tasksQuery task contracts — filter by realm, session, status, or tagvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max results (default 50)
realm	string	No	—	Filter by realm
session_id	string	No	—	Filter by session ID
status	any	No	—	Filter by status (0-4)
tag	string	No	—	Filter by tag

register_taskRegister a task contract — records goal, constraints, acceptance criteria, priority, and optional deadline for an ongoing agent taskvia advanced

Parameter	Type	Required	Default	Description
acceptance_criteria	array<string>	No	—	Criteria for task completion
constraints	array<string>	No	—	Constraints that must be respected
deadline_ms	any	No	—	Optional deadline as Unix ms timestamp
goal	string	Yes	—	Task goal description
parent_task_id	any	No	—	Parent task ID for subtasks
priority	any	No	—	Priority 1-10 (default 5)
realm	string	No	—	Realm: coding, research, planning (default: coding)
session_id	string	No	—	Session this task belongs to
tags	array<string>	No	—	Optional tags

resolve_probeResolve a pending probe — mark as Answered (1) or Dismissed (2) and optionally record the answervia advanced

Parameter	Type	Required	Default	Description
answer	string	No	—	Optional answer text
probe_id	any	Yes	—	Probe ID
status	any	Yes	—	1=Answered 2=Dismissed

set_criterionUpsert a completion criterion for a task — creates if new, updates if existing criterion text matchesvia advanced

Parameter	Type	Required	Default	Description
criterion	string	Yes	—	Criterion description
evidence_note	string	No	—	Optional evidence supporting the criterion check
is_met	boolean	No	—	Whether criterion is met (default false)
task_id	any	Yes	—	Task ID

skill_deprecateDeprecate a skill (marks latest version as deprecated).gateway, via advanced

Parameter	Type	Required	Default	Description
skill_id	string	Yes	—	Skill identifier to deprecate

skill_listList all registered skills with their latest version number.gateway, via advanced

skill_readRead a skill version. version=0 means latest.gateway, via advanced

Parameter	Type	Required	Default	Description
skill_id	string	Yes	—	Skill identifier
version	integer	No	—	Version number (0=latest)

skill_searchSearch skills by text match on id, tags, or content.gateway, via advanced

Parameter	Type	Required	Default	Description
limit	integer	No	—	Max results (default: 20)
query	string	Yes	—	Search query

skill_uploadUpload a new version of a reusable skill. Skills are immutable versioned text blobs (prompts, templates, procedures).gateway, via advanced

Parameter	Type	Required	Default	Description
content	string	Yes	—	Skill content (prompt, template, procedure)
skill_id	string	Yes	—	Unique skill identifier
tags	array<string>	No	—	Tags for search
uploaded_by	string	No	—	Agent or user who uploaded

update_taskUpdate task status (Active=0, Blocked=1, Completed=2, Failed=3, Abandoned=4), optionally attach an intervention ID or add a tagvia advanced

Parameter	Type	Required	Default	Description
add_intervention_id	any	No	—	Attach intervention to task
add_tag	string	No	—	Add a tag to the task
status	any	Yes	—	0=Active 1=Blocked 2=Completed 3=Failed 4=Abandoned
task_id	any	Yes	—	Task ID

Intervention Ledger

8 tools ▾

add_observationRecord an observation during an open intervention (stdout, test result, file diff, etc.)via advanced

Parameter	Type	Required	Default	Description
confidence	number	No	—	Confidence in this observation (0.0-1.0)
evidence_refs	array<any>	No	—	Memory IDs that constitute evidence
intervention_id	any	Yes	—	Intervention ID from start_intervention
kind	any	No	—	0=Stdout 1=Stderr 2=FileDiff 3=TestResult 4=EnvState 5=UserFeedback
summary	string	Yes	—	Human-readable observation summary

close_interventionClose an intervention with its final outcome statusvia advanced

Parameter	Type	Required	Default	Description
intervention_id	any	Yes	—	Intervention ID to close
status	any	Yes	—	0=Open 1=Succeeded 2=Failed 3=Partial 4=Aborted

get_interventionGet a single intervention record by IDvia advanced

Parameter	Type	Required	Default	Description
intervention_id	any	Yes	—	Intervention ID

intervention_statsShow intervention ledger statistics — total, open, succeeded, failed, aborted countsvia advanced

list_open_interventionsList all currently open (in-progress) interventionsvia advanced

query_interventionsQuery the intervention ledger with optional filtersvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max results (default 50)
realm	string	No	—	Filter by realm
session_id	string	No	—	Filter by session ID
status	any	No	—	Filter by status (0-4)

record_attributionAttribute a closed intervention to a causal class — routes feedback to the appropriate learning subsystemvia advanced

Parameter	Type	Required	Default	Description
confidence_delta	number	No	—	Magnitude of the learning signal (0.0-1.0)
debt_ids	array<any>	No	—	Linked epistemic debt IDs
intervention_id	any	Yes	—	Intervention ID
note	string	No	—	Optional human-readable note
primary_class	any	Yes	—	0=MemoryRecallError 1=SourceTrustError 2=ProcedureError 3=ToolExecutionError 4=EnvironmentShift 5=HiddenPrecondition 6=AmbiguousState 7=GoalSpecError 8=UserOverride 9=ExternalNondeterminism
secondary_class	any	No	—	Optional secondary attribution class (same enum)
skill_memory_ids	array<any>	No	—	Skill memory IDs that were applied
source_memory_ids	array<any>	No	—	Memory IDs that contributed to this outcome
surprise_id	any	No	—	Linked surprise event ID if available

start_interventionBegin tracking an agent intervention — records intent, action, preconditions and expected observables before executionvia advanced

Parameter	Type	Required	Default	Description
action_ref	string	Yes	—	Reference to the action (tool name, file path, command)
action_type	any	No	—	0=ToolCall 1=MultiStepPlan 2=Delegation 3=Edit 4=Command
agent_id	string	No	—	Agent performing the action
domain	string	No	—	Domain area (e.g. git, filesystem, testing, compiler)
expected_observables	array<string>	No	—	What success looks like
intent	string	Yes	—	What the agent intends to achieve
preconditions	array<string>	No	—	Known preconditions
realm	string	No	—	Realm: coding, research, planning (default: coding)
reversal_cost	any	No	—	0=None 1=Low 2=Medium 3=High
session_id	string	No	—	Current session ID
task_id	any	No	—	Optional task ID

Executable Constraints

7 tools ▾

assert_factAssert a constraint fact (subject-predicate-object) with provenance and scope. Auto-detects conflicts and creates rival branches.via advanced

Parameter	Type	Required	Default	Description
branch_id	any	No	—	Branch to assert into (0=trunk)
confidence	number	No	—	Confidence 0-1 (default 0.8)
confidence_basis	string	No	—	Basis: stated, observed, derived, corrected
object	string	Yes	—	Value (e.g. 'Rust', 'vim', 'Copenhagen')
predicate	string	Yes	—	Relation (e.g. 'prefers', 'uses', 'located-in')
provenance_source	string	No	—	Source: user, tool, distillation, inference
scope	string	No	—	Scope: global, realm name, or session (default: global)
subject	string	Yes	—	Entity (e.g. 'user', 'project-X')

branch_createFork a rival branch for conflicting interpretationsvia advanced

Parameter	Type	Required	Default	Description
parent_id	any	No	—	Parent branch ID (0=trunk)
scope	string	No	—	Branch scope (default: global)

branch_resolveResolve a branch conflict: winner stays, loser abandonedvia advanced

Parameter	Type	Required	Default	Description
loser_id	any	Yes	—	Branch ID to abandon
winner_id	any	Yes	—	Branch ID that wins

explain_factExplain a fact: provenance chain + supporting/conflicting factsvia advanced

Parameter	Type	Required	Default	Description
fact_id	any	Yes	—	Fact ID to explain

query_chainFollow predicate chain: A→B→C through constraint factsvia advanced

Parameter	Type	Required	Default	Description
predicates	array<string>	Yes	—	Ordered list of predicates to follow
subject	string	Yes	—	Starting entity

query_unifyPattern-match query against constraint store (unification with wildcards)via advanced

Parameter	Type	Required	Default	Description
object	string	No	—	Object filter
predicate	string	No	—	Predicate filter
scope	string	No	—	Scope filter
subject	string	No	—	Subject filter (omit for wildcard)

retract_factSoft-retract a constraint fact (preserves history)via advanced

Parameter	Type	Required	Default	Description
fact_id	any	Yes	—	Fact ID to retract

Surprise & Epistemic Debt

11 tools ▾

attach_debt_evidenceAttach supporting evidence to an epistemic debt — memory IDs + confidencevia advanced

Parameter	Type	Required	Default	Description
confidence	number	No	—	Evidence confidence 0-1 (default 0.5)
debt_id	any	Yes	—	Epistemic debt ID
memory_ids	array<any>	No	—	Memory IDs that serve as evidence
note	string	No	—	Optional note about the evidence

debt_statsSummary statistics for epistemic debt: counts by status, avg fragilityvia advanced

defer_debtDefer an epistemic debt for later investigationvia advanced

Parameter	Type	Required	Default	Description
debt_id	any	Yes	—	Debt ID to defer

get_blind_spotsIdentify recurring surprise patterns — domains/actions where predictions consistently failvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max blind spots (default 10)
realm	string	No	—	Filter by realm

get_fragile_decisionsList open epistemic debts sorted by fragility — decisions most likely to be wrongvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—	Max results (default 20)
threshold	number	No	—	Minimum fragility threshold (default 0.5)

query_debtsQuery epistemic debts with filtersvia advanced

Parameter	Type	Required	Default	Description
domain	string	No	—	Filter by domain
limit	any	No	—	Max results (default 50)
min_fragility	number	No	—	Minimum fragility score
realm	string	No	—	Filter by realm
status	string	No	—	Filter: open, resolved, deferred

query_surprisesQuery recorded surprise/prediction-error events with filtersvia advanced

Parameter	Type	Required	Default	Description
domain	string	No	—	Filter by domain
limit	any	No	—	Max results (default 50)
min_magnitude	number	No	—	Minimum surprise magnitude
realm	string	No	—	Filter by realm
since_ms	any	No	—	Only events after this timestamp (ms)

record_surpriseRecord a prediction error event — what was expected vs what actually happenedvia advanced

Parameter	Type	Required	Default	Description
action	string	No	—	What action was taken
actual	string	Yes	—	What actually happened (required)
context_sketch	string	No	—	What was happening when the surprise occurred
domain	string	No	—	Domain: recall, tool, user_correction, constraint (default general)
expected	string	No	—	What was predicted/expected (optional)
realm	string	No	—	Realm filter (default global)
session_id	string	No	—	Session ID (optional)
source_memory_id	any	No	—	Related memory ID (optional)
surprise_magnitude	number	No	—	How surprising [0-1] (default 0.5)

register_debtRegister an epistemic uncertainty — competing hypotheses that need resolutionvia advanced

Parameter	Type	Required	Default	Description
competing_hypotheses	array<string>	No	—	Competing explanations
discriminating_test	string	No	—	How to distinguish between hypotheses
domain	string	No	—	Domain (default general)
fragility_score	number	No	—	How fragile this belief is [0-1] (default 0.5)
pattern	string	Yes	—	The uncertain pattern/belief (required)
realm	string	No	—	Realm (default global)
session_id	string	No	—	Session ID (optional)

resolve_debtMark an epistemic debt as resolved with a resolutionvia advanced

Parameter	Type	Required	Default	Description
debt_id	any	Yes	—	Debt ID to resolve
resolution	string	Yes	—	How the uncertainty was resolved

surprise_statsSummary statistics for surprise memory: counts, avg magnitude, domain breakdownvia advanced

Wisdom Lifecycle

14 tools ▾

close_rederiveClose a re-derivation contract for an Inflamed wisdom lineage. Actions: reaffirm (0), narrow (1), split (2), demote (3).via advanced

Parameter	Type	Required	Default	Description
action	any	Yes	—	0=reaffirm 1=narrow 2=split 3=demote
fork_claim	string	No	—	Claim for the forked lineage (action=split)
fork_lineage_id	any	No	—	Pre-enrolled fork lineage ID (action=split)
lineage_id	any	Yes	—	Lineage ID
new_envelope	object	No	—	Narrowed applicability envelope (for action=narrow/split)

enroll_wisdom_lineageEnroll a Trusted wisdom candidate into the Wisdom Homeostasis layer — creates a living WisdomLineage record that tracks belief integrity over timevia advanced

Parameter	Type	Required	Default	Description
ancestor_lineage_id	any	No	—	Parent lineage ID if this is a fork/split
claim	string	Yes	—	The claim this wisdom encodes
derivation_relation	string	No	—	Relation to ancestor: supersedes\|branches_from\|narrows\|splits_from
envelope	object	Yes	—	Applicability envelope: {domain, action_types, preconditions, source_families}
seed_debt_ids	array<any>	No	—	Epistemic debt IDs
seed_episode_ids	array<any>	No	—	Episode IDs that seeded this wisdom
seed_intervention_ids	array<any>	No	—	Intervention IDs
seed_surprise_ids	array<any>	No	—	Surprise event IDs
wisdom_candidate_id	any	Yes	—	ID of the WisdomCandidate to enroll

get_wisdom_lineageGet full details of a wisdom lineage by ID, including challenger evidence and state historyvia advanced

Parameter	Type	Required	Default	Description
lineage_id	any	Yes	—	Lineage ID

insight_globalList global insightsvia advanced

Parameter	Type	Required	Default	Description
limit	any	No	—
tag	string	No	—

insight_promotePromote memory to globalvia advanced

Parameter	Type	Required	Default	Description
id	any	Yes	—
reason	string	No	—

lineage_expiry_checkList Inflamed lineages whose re-derivation TTL has expired — these should be demoted or re-derived urgentlyvia advanced

query_wisdom_candidatesQuery wisdom candidates by lifecycle stage and/or domainvia advanced

Parameter	Type	Required	Default	Description
domain	string	No	—	Filter by domain
lifecycle	any	No	—	Filter by lifecycle: 0=candidate, 1=provisional, 2=trusted, 3=demoted
limit	any	No	—	Max results (default 50)

query_wisdom_lineagesList wisdom lineages filtered by state and/or domainvia advanced

Parameter	Type	Required	Default	Description
domain	string	No	—	Filter by domain
limit	any	No	—	Max results (default 50)
state	string	No	—	Filter: trusted\|watch\|inflamed\|demoted

tick_lineage_stalenessManually trigger a staleness tick — grows staleness mass on lineages with no recent support. Normally called by the subconscious cycle.via advanced

transition_wisdom_lineageManually transition a wisdom lineage state (Trusted/Watch/Inflamed/Demoted). Normally automatic — use for overrides.via advanced

Parameter	Type	Required	Default	Description
lineage_id	any	Yes	—	Lineage ID
new_state	any	Yes	—	0=Trusted 1=Watch 2=Inflamed 3=Demoted
reason	string	No	—	Why this transition is happening
rederive_task_id	any	No	—	Task contract ID if opening re-derivation

update_wisdom_lifecycleAdvance a wisdom candidate through lifecycle stages: candidate→provisional→trusted→demotedvia advanced

Parameter	Type	Required	Default	Description
candidate_id	any	Yes	—	Wisdom candidate ID
new_state	any	Yes	—	0=candidate, 1=provisional, 2=trusted, 3=demoted

upsert_wisdom_candidateCreate or update a wisdom candidate from clustered surprise patternsvia advanced

Parameter	Type	Required	Default	Description
action	string	No	—	Action or behavior pattern
cluster_key	string	Yes	—	Unique key for this pattern cluster (domain+action+sig)
cross_session_count	any	No	—	Number of distinct sessions with evidence
debt_ids	array<any>	No	—	Resolved debt IDs linked to this candidate
domain	string	No	—	Knowledge domain
episode_ids	array<any>	No	—	Surprise event IDs supporting this candidate
mean_surprise	number	No	—	Average surprise magnitude across episodes
promotion_score	number	No	—	Computed promotion readiness score 0-1
summary	string	No	—	Human-readable summary of the wisdom
support_count	any	No	—	Number of supporting episodes

wisdom_lineage_statsShow wisdom homeostasis statistics — counts by state, mean staleness, support/contradiction mass totalsvia advanced

wisdom_promotion_statsOverview of wisdom promotion pipeline — total candidates by lifecycle stagevia advanced

Causal Episode Compiler

5 tools ▾

executor_flushPromote shadow intervention policies that have passed the 20-event / lift>0.15 gate, demote policies whose source rule is refuted, and report active policy stats. Safe to call any time; idempotent.via advanced

hypothesis_probesTop-k Sequitur rules ranked by expected information gain (Wilson probe_value). Maximized at p_hat=0.5 — rules the system is most uncertain about. Run consolidation_pass first to populate.via advanced

Parameter	Type	Required	Default	Description
k	any	No	—	Max rules to return (default 10)

log_eventLog a structured action event to the CEC tape and CDAWG (tool, entity, outcome: 0=success 1=fail 2=error 3=partial)via advanced

Parameter	Type	Required	Default
entity	string	Yes	—
outcome	any	No	—
session_id	any	No	—
tool	string	Yes	—
ts_ms	any	No	—

log_event_exLog a CEC event with regret-shaping telemetry (token_cost, latency_ms, retry_count). Updates Q-values with utility = outcome_reward - 0.001*token_cost - 0.00001*latency_ms - 0.1*retry_count.via advanced

Parameter	Type	Required	Default	Description
entity	string	Yes	—
latency_ms	any	No	—	Wall-clock ms (0=unknown)
outcome	any	No	—	0=success 1=fail 2=error 3=partial
retry_count	any	No	—	Retries before this outcome
session_id	any	No	—
token_cost	any	No	—	Tokens consumed (0=unknown)
tool	string	Yes	—
ts_ms	any	No	—

refutation_statsShow Sequitur rules that are being falsified: rules whose antecedent appears but is NOT followed by the expected consequent. Returns live/refuted counts and top-k by refutation ratio.via advanced

Parameter	Type	Required	Default	Description
k	any	No	—	Max rules to show (default 10)

Import / Export & Files

5 tools ▾

advancedGateway to hidden/advanced tools. Use action='list' to see every hidden tool (about 250), or call directly with tool='<name>' and arguments={...}. Example: {"tool": "pin_memory", "arguments": {"id": 123}}gateway

Parameter	Type	Required	Default	Description
action	string	No	—	Action: 'list' to show available tools
arguments	object	No	—	Arguments to pass to the hidden tool
category	string	No	—	Filter list by category: 'advanced' or 'internal'
tool	string	No	—	Hidden tool name to call

file_at_timeGet file content at time (stub)via advanced

Parameter	Type	Required	Default
file_path	string	Yes	—
session_id	string	No	—
show_diff	boolean	No	—
time	string	No	—

file_restoreRestore file version (stub)via advanced

Parameter	Type	Required	Default
file_path	string	Yes	—
preview	boolean	No	—
version_id	any	No	—

file_timelineShow files modified in time rangevia advanced

Parameter	Type	Required	Default
cross_session	boolean	No	—
file_pattern	string	No	—
limit	any	No	—
path	string	No	—
query	string	No	—
session_id	string	No	—

run_hint_enricherGenerate retrieval hints for unprocessed memories using chitta-hint-tuned. Reads memories without a retrieval_hint tag, calls the local hint model, stores each hint as a derived memory (kind=hint, tags=retrieval_hint), and marks the source memory with hint:done. Run after a session to enrich new memories for better recall.gateway

Parameter	Type	Required	Default	Description
dry_run	boolean	No	false	Preview hints without writing to memory (default: false)
limit	integer	No	100	Max memories to enrich per run (default: 100)
model	string	No	"chitta-hint-tuned"	Ollama model to use (default: chitta-hint-tuned)

Other

1 tool ▾

prompt_contextApply shared prompt admission and render its budget metadatavia advanced

Parameter	Type	Required	Default	Description
local	boolean	No	—	CLI only: run the same pure policy locally for timeout fallback
state	object	Yes	—	Lane text, session hashes, query tokens and policy settings

Unadvertised native handlers

These registered daemon handlers have no advertised MCP schema. They remain callable by name or through advanced. Arguments are validated by the daemon; absence of a schema does not mean no arguments.

anticipation_gate_status (hidden)
anticipation_record_outcome (hidden)
batch_forget (hidden)
chitta_health (hidden)
cleanup (hidden)
clear_codebase (hidden)
clear_triplets (hidden)
consolidation_auto (hidden)
consolidation_merge (hidden)
consolidation_scan (hidden)
cycle (hidden)
dedupe_symbols (hidden)
describe_symbol (hidden)
distill_status (hidden)
enrichment_status (hidden)
epiplexity_check (hidden)
export_soul (hidden)
export_training_pairs (advanced)
extract_symbols (hidden)
fep_status (advanced)
file_dependents (hidden)
file_imports (hidden)
file_index_all (hidden)
file_index_session (advanced)
harvest_scope (advanced)
health_check_start (advanced)
hygiene_run (hidden)
import_soul (hidden)
ingest_source (advanced)
ledger_append (advanced)
ledger_compile (advanced)
ledger_contradictions (advanced)
ledger_op (advanced)
ledger_query (advanced)
metacognition_corrections (hidden)
metacognition_outcomes (hidden)
msg_ack (hidden)
predicate_attach (advanced)
predicate_list (advanced)
predicate_run (advanced)
queue_experiments (advanced)
recall_temporal_events (advanced)
reconcile_pass (advanced)
reembed_memories (hidden)
repl_execute (advanced)
repl_session_delete (advanced)
repl_session_get (advanced)
repl_session_list (advanced)
repl_session_set (advanced)
resolve_callsites (hidden)
restore_code_intel_confidence (hidden)
routed_recall (advanced)
seed_hdc_geometry (advanced)
session_deregister (hidden)
session_heartbeat (hidden)
session_register (hidden)
ssl_convert (hidden)
subconscious_stats (hidden)
suggestion_count (hidden)
suggestion_pending (hidden)
suggestion_resolve (hidden)
suggestion_track (hidden)
tape_stats (advanced)
transcript_get (hidden)
transcript_list (hidden)
transcript_parse (hidden)
transcript_register (hidden)
transcript_remove (hidden)
transcript_search (hidden)
transcript_update (hidden)
turiya_status (advanced)
type_hierarchy (hidden)
verbalize_rules (advanced)
version_check (hidden)
wiki_export (advanced)
witness_memory (advanced)

← Back to chitta