chitta — Philosophy

Vedantic Concepts Memory Architecture Retrieval Learning Processing Integration

Part I

Vedantic Foundations

Each Sanskrit term names a measurable operation

ब्रह्मन् Brahman Universal soul

→ Shared memory field across all sessions. The default realm where memories live unless scoped to a project. Global visibility makes records available across realms.

आत्मन् Ātman Individual self

→ Per-project identity and preference state. Each realm carries its own memories, decisions, and calibration. The self that emerges through one project differs from another.

अन्तःकरण Antahkarana Inner instrument

→ Six cognitive voices as memory channels. Manas captures, Buddhi discerns, Ahamkara personalizes, Chitta stores, Vikalpa generates alternatives, Sakshi witnesses.

चित्त Chitta Memory store

→ Persistent embeddings and semantic indexing in a Rust store behind the C++ daemon. Vector dimension follows the compiled model identity: 768 on the measured deployment, 1024 in public defaults.

यन्त्र Yantra Instrument

→ VakYantra (speech instrument) transforms text into embeddings. Not passive encoding — the model shapes what becomes memorable by how it embeds.

अनित्य Anitya Impermanence

→ Decay curves and time-weighted forgetting. Memories are not permanent. Confidence decays over time. Unused memories fade. Decay reduces confidence in unused records.

संस्कार Saṃskāra Impressions

→ Hebbian²⁶ learning and usage-weighted strengthening. When a memory is recalled and used, it grows stronger. Frequent patterns crystallize into wisdom.

प्रत्यभिज्ञा Pratyabhijñā Recognition

→ Resonance-based recall ranks stored memories against the current context.

यज्ञ Yajña Sacred offering

→ Background compression and wisdom extraction. Transcript distillation extracts condensed records from episodes.

Part II

Memory Architecture

Why we store what we store, how we store it

Memory storage and retrieval

The store holds records and the indexes used to retrieve them. Index structure and scoring determine which records are returned.

Connectionism research Harder to debug

Embedding and payload storage

Dense embeddings and payloads use separate structures and sidecars within one Rust store. Search indexes are derived state; snapshot families and the WAL preserve authoritative memory.

Database sharding Join complexity

chitta-field → sparse codes (64/16K) + cortical posting index WAL segments → content, metadata, relationships, durable log

Directional relationships

Semantic similarity finds related concepts. But "X caused Y" or "user prefers A over B" needs explicit relationships. Triplets encode directional knowledge.

Knowledge graphs Storage overhead

Theme clustering

No predefined categories. Themes emerge from embedding clusters using HDBSCAN. The system discovers what concepts naturally group together based on your usage.

xMemory paper Unpredictable categories

Confidence decay floor

Memory confidence decreases through exponential decay, but asymptotically approaches a floor (0.1). Old memories fade but never fully disappear.

Ebbinghaus forgetting curve Unbounded storage

new_confidence = floor + (current - floor) * 0.995 Decay every 60s during idle. Floor = 0.1.

SSL compression

Soul Shorthand Language condenses records before embedding. The documented token reduction is 40-60%; retrieval depends on the encoded content and embedding model.

Compression benchmarks Learning curve

[category] Title - Detail (context) → Implication

Part III

Retrieval Philosophy

How memories surface at the right moment

Hybrid retrieval

Retrieval combines vector similarity, BM25²⁴ keyword matching, and graph traversal.

HyDE paper, RAG research Complexity, latency

Context budget

Claude's ~200K token context is finite. Every surfaced memory consumes space. Retrieval maximizes signal-to-noise, not total recall.

LLM attention research Useful memories filtered

Session start: ~2000 tokens Per-query recall: ~500 tokens Correction warnings: ~100 tokens

Part IV

Learning Mechanisms

How the system adapts to you

Correction records

Corrections are captured automatically, stored with boosted confidence, and retrieved as warnings about recorded mistakes.

Error-driven learning False positive corrections

Embedding model

VakYantra converts text into embedding vectors. The model determines the vector space used in retrieval.

Embedding model research Model-locked

Part V

Processing Model

Background operations and lifecycle

Hook capture

Hooks capture session boundaries, prompt submissions, and tool calls.

Lifelogging research Privacy surface

Background processing

The daemon runs background tasks during idle: confidence decay, theme clustering, memory consolidation. These would interrupt conversation if done synchronously.

OS scheduling, spaced repetition Debugging difficulty

Part VI

Integration Design

How chitta fits into Claude Code

Resident daemon

A persistent daemon (chittad) keeps the chitta-field memory substrate and ONNX model warm in memory. The documented model cold-start cost is approximately 750 ms.

Connection pooling, model serving Memory footprint (~1GB)

Unix socket IPC. Thread pool (8-16 workers). Query latency depends on cache and index state.

These concepts supply names for memory storage, decay, reinforcement, and retrieval. The implementation and measurements describe their behavior.

Full claim files with cross-references available in the philosophy/ directory.

← Back to chitta | Architecture →

References

Stephen Robertson and Hugo Zaragoza. The Probabilistic Relevance Framework: BM25 and Beyond. Foundations and Trends in Information Retrieval 3(4), 333–389 (2009). https://doi.org/10.1561/1500000019
Donald O. Hebb. The Organization of Behavior: A Neuropsychological Theory. Wiley (1949); Psychology Press reissue (2002). https://www.routledge.com/The-Organization-of-Behavior-A-Neuropsychological-Theory/Hebb/p/book/9780415654531