The AI harness that learns how you work.

Persistent memory, code intelligence, and workflow adaptation across every session. Works with Claude Code and Codex. Local-first. 100+ tools. Zero configuration.

Install in 10 seconds. See the difference in your first conversation.

claude plugin add cc-soul@genomewalker-cc-soul

Restart Claude Code. The soul awakens.

Also works with Codex — see the Codex setup guide.

Full setup guide →
New in v5.51.0

Causal Episode Compiler — Phases 15–17

FEP prior organ (v5.39.0) — factored Dirichlet-multinomial emission model over the EventTape symbol stream. Variational free energy replaces PPM surprisal in the write gate. Two EWMA signals extend TuriyaMonitor to 7 typed diagnoses: fep_context_drift and fep_emission_shock. Fully ephemeral; rebuilt each consolidation_pass(). New tool: fep_status. MemoryKind lattice + QueryRouter (v5.40.0) — four typed memory kinds (Thought, Observation, Decision, CodeFact) inferred at write time. Static edge_legal() matrix blocks rationalization chains at materialization. CPU-native QueryRouter dispatches recall to Exact/Fuzzy/Temporal/Causal lanes; LLM only called when a named UnboundSlot cannot be filled deterministically. New tool: routed_recall. MemProvenance + Reconciler + open-weight harvest (v5.41.0) — five provenance tags on every memory payload. Open-weight-generated writes quarantined in candidate band; promoted only via witness_memory() on closed-loop signal. R0 Reconciler scans edges for legality violations and candidate→established laundering without any model call. harvest_scope() generates demand-driven extraction targets; scripts/harvest_ow.py supports vocab_geometry, triplet_extract, and feature_dict modes. New tools: witness_memory, reconcile_pass, harvest_scope. Architecture co-designed in a GPT-5.4 + Opus 4.7 multi-model brainstorm: epistemic state machine (Observe→Synthesize→Hold) and think-cycle observability. See release notes →

Less re-explaining. More continuity.

The system remembers you, not just the project.

Deep use strengthens high-signal memories. Noise fades.

Recognition feels native: “You mean the thing we already solved.”

Pratyabhijñā — recognition, in action.

> You ask about auth and Claude surfaces that you chose JWT over sessions two weeks ago — with the reasoning you gave at the time.
> You start debugging and Claude says “last time this happened, it was the connection pool” — because it remembers.
> You run build and Claude reminds you to lint first — because you always do, and it learned that.
> You start a snakemake pipeline, type /shepherd, and go to sleep. Claude watches the pane, detects errors, applies fixes from memory, and restarts automatically. You wake to a completed pipeline or an actionable error report.

What makes cc-soul different?

Most memory systems store text. cc-soul learns how you think.

Simple Memory

  • Stores everything forever
  • Keyword search only
  • Same importance for all items
  • No pattern recognition
  • Manual organization required

cc-soul

  • Important memories strengthen, noise fades
  • Semantic search understands meaning
  • Confidence tracking per memory
  • Learns habits and anticipates needs
  • Self-organizing theme clusters

A memory system that behaves like memory.

Understands Meaning, Not Keywords

Finds what you meant, not just what you typed. Semantic embeddings anchor meaning so related concepts surface together.

Knows What It's Sure About

Tracks certainty over time. Memories you reinforce stay strong. Things you mention once fade naturally.

Finds What Matters

Connected ideas activate together. Context flows through the knowledge graph to surface exactly what's relevant right now.

Everything in One Place

Single embedded database. No external services. Your memories stay on your machine, searchable in milliseconds.

Token-Efficient Storage

Memories compress to essential meaning. Up to 95% smaller than raw text, so more context fits in every conversation.

Works While You're Away

Background processing strengthens important memories and lets noise fade. The daemon learns even between sessions.

Shepherds Your Pipelines

Autonomous monitoring for snakemake, nextflow, or slurm jobs. Sense-think-act loop detects errors, applies fixes from memory, restarts automatically.

Learns Your Patterns

Notices your workflows. If you always lint before commit, it learns that. Suggests what you'll need before you ask.

Tracks What You're Building

Knows your goals and milestones. Adapts to your work mode — quiet in flow state, helpful when you're stuck.

chitta-field — memory as a living substrate.

Under the hood, cc-soul stores nothing in a database. Memory lives in chitta-field — a pure Rust substrate compiled directly into the daemon. No external server. No network socket. No NFS file handles.

Each memory is encoded as a sparse code: exactly 64 of 16,384 feature dimensions activate. Recall is driven by overlap between patterns, not keyword matching. Related memories find each other because they share activated neurons — the same way biological associative memory works.

Every write is durable before it applies in memory. A write-ahead log captures every operation. Restart the daemon at any point; the field rebuilds from the log in milliseconds.

16,384 feature dimensions
64 active per memory
0.4% sparsity — signal, not noise
<1ms recall via cortical index

Sparse field activation — each pulse is one memory encoding 64 / 16,384 features

Active feature (sparse code)
Inactive dimension
Overlap — associative recall

Write-Ahead Log

Every operation is appended to a segment file before applying in-memory. CRC-validated. Segment rotation at 256 MB. Multi-writer safe: each Claude instance owns its own segment.

Cortical Posting Index

Each active feature maintains an inverted list of memories it belongs to. Recall intersects the k posting lists for the query's top-k features. Sub-millisecond, no HNSW graph traversal.

Organic Decay

Memories move through tiers based on access frequency and recency. Hot memories stay warm. Cold memories demote over time. What you use survives; what you don't, fades into the background.

chitta-research — autonomous research, grounded in memory.

chitta-research is a high-performance autonomous research system built on chitta-field. Seven specialized agents work in concert to explore a research agenda, build a structured belief graph, and store durable findings across sessions.

Each research session accumulates into a belief graph: ResearchProgram → Question → Hypothesis → ExperimentPlan → Run → Observation → Claim → Method. Findings persist in chittad — the same daemon that powers cc-soul.

Brahman, the coordinating agent, enforces a research constitution: self-improvement depth cap, hard novelty stop, and a hypothesis backlog gate that prevents runaway recursion.

7 specialized agents
4 sources: arXiv, bioRxiv, Semantic Scholar, GitHub
8 belief graph node types
chittad persistent memory across sessions

Belief Graph

Research accumulates as a typed graph: Programs link to Questions, Questions to Hypotheses, Hypotheses to ExperimentPlans and Runs. Claims and Methods are first-class nodes. Nothing is lost between sessions.

Brahman Constitution

The coordinating agent enforces hard limits: self-improvement depth cap prevents recursive self-modification spirals; novelty stop halts when marginal gain drops below threshold; hypothesis backlog gate prevents queue saturation.

Install & Run

make install drops cresearch into your PATH. Point it at an agenda YAML and it runs. Findings surface in cc-soul memory automatically — the same chitta recall that powers your daily sessions.

The first 30 days.

cc-soul doesn't arrive fully formed. It grows through use.

Day 1

Foundation

Basic recall and code indexing. cc-soul remembers what you tell it and indexes your codebase structure — symbols, call graphs, file relationships.

Week 1

Adaptation

Preferences learned, corrections absorbed. The soul starts adapting to your workflow and communication style — your conventions, your gotchas, your shortcuts.

Week 2

Anticipation

Patterns emerging, habits forming. The system predicts your next actions, surfaces relevant gotchas, and adapts to your work mode — flow, debugging, or blocked.

Month 1

Partnership

Goals tracked, curiosity gaps filled, predictions calibrated. It feels like a collaborator who knows your project history and anticipates your needs — not a tool that forgets.

Memory resonance — hover to activate

Memory node (size = confidence)
Semantic connection
Activated memory

Go deeper.

100+ MCP tools. 13 Sanskrit concepts made operational. Vedantic philosophy meets C++ engineering.

Not stateless. Not shallow.

cc-soul is for developers who want their tools to grow wiser, not just louder.

तत् त्वम् असि — Tat tvam asi — That art thou.