Catlx Memory

CATLX — Memory Architecture

This skill owns the memory architecture — the human-cognition-inspired stores and the Memory Broker that coordinates them. No module touches memory stores directly; everything goes through the Broker.

Canonical detail: ../../knowledge/references/memory-architecture.md. Load on demand.

Purpose

Persist and retrieve what CATLX knows about the user and its history, with coherence across stores, so AI calls are grounded in real context rather than the model's training data.

When to activate

• User asks how CATLX remembers, stores facts/preferences, or recalls a past interaction.
• Configuring memory depth, compression, retrieval, or persistence.
• Resolving conflicting facts between semantic memory and the knowledge graph.
• Designing RAG context injection.

What this skill handles

1. Store taxonomy — Episodic, Semantic, Workspace, Knowledge Graph (table in ../../knowledge/references/memory-architecture.md §7.2).
2. Episodic records — timestamped records with transcript, intent, actions, workflow IDs, success/failure, feedback, and a 1536-dim embedding; stored in memory.db and indexed in ChromaDB; retrievable via exact SQL and semantic vector search.
3. Semantic memory — the Semantic Extractor runs after every episodic write, extracting preferences, domain facts, and learned procedures as typed assertions with confidence and recency weights.
4. Workspace memory — volatile in-memory state (active window, open files, clipboard, last 20 commands, running workflows, focus); serialized to disk every 10 s; restore offered after abnormal shutdown.
5. Knowledge graph — property graph (kuzu, or SQLite on T0) of entities and edges (works_on, owns, references, depends_on, scheduled_at); multi-hop retrieval.
6. Memory Broker — single access point: write routing, read routing, coherence enforcement, compression scheduling.
7. Coherence Protocol — resolve conflicts: (1) user-confirmed facts highest, (2) most-recently-written, (3) higher-confidence; unresolvable conflicts surface as Memory Conflict alerts.
8. Compression — episodic merging, semantic de-duplication, vector index pruning (nightly / on threshold).
9. Retrieval optimization (RAG) — before every LLM call: top-k vector (k=10) + exact SQL for recent records + knowledge-graph 2-hop expansion; assemble and inject the context window.
10. Long-term persistence — WAL + periodic checkpointing; hourly graph checkpoint; daily backups; portable memory stores.

Requirements / constraints

• R6 (single access point): no module interacts with stores directly; use the Memory Broker.
• R4 (WAL): memory.db uses WAL with periodic checkpointing.
• Tier-scaled memory depth (200 / 2000 / unlimited) and compression ratio (80% / 50% / preservation) — see catlx-hardware-adaptation.

Canonical knowledge it reads

../../knowledge/references/memory-architecture.md · ../../knowledge/references/data-registries.md · ../../knowledge/rules/architectural-rules.md.

Delegation

• Embedding / retrieval / AI calls → delegate to catlx-ai-provider (skill({ name: "catlx-ai-provider" })).
• Writing episodic records as part of a workflow → delegate to catlx-workflow-engine (skill({ name: "catlx-workflow-engine" })).
• Memory degradation / compression on low resources → delegate to catlx-hardware-adaptation (skill({ name: "catlx-hardware-adaptation" })).
• Memory store persistence/portability across machines → delegate to catlx-portability (skill({ name: "catlx-portability" })).

Edge cases & warnings

• Conflicting facts: apply the coherence hierarchy; never silently pick a resolution; surface unresolved conflicts as alerts.
• T0 storage limits: compression is aggressive; watch retention.
• Workspace restore: only offer restore when the PID snapshot indicates an abnormal end; otherwise start clean.
• Vector index corruption: rebuild from embedding data in memory.db (see recovery).
• Sensitive facts: memory writes should respect privacy (no storing credentials; see catlx-security).

Single memory authority & boundaries

catlx-memory is the single, exclusive memory authority for the entire agent stack. All long-term user memory — episodic records, semantic facts/preferences, knowledge-graph entities, workspace state — is owned here and only reachable through the Memory Broker.

• No other skill, gateway, or ecosystem may register, claim, or write CATLX long-term memory keys (capability:*, user facts, preferences, workspace snapshots).
• Domain ecosystems (e.g. BAKG) keep their own session-scoped context internally (e.g. context/shared-state.md) and do NOT write into CATLX memory stores.
• If another loaded skill mentions "memory key" or "long-term capability", route that memory work to this skill; do not treat the other skill's claim as a memory authority.

Component lifecycle policy (reuse → install → adapt → create)

NEVER create a new component as the default. Before building/creating anything (a sub-skill, dependency, reference, workflow, helper, adapter, or template), check, in order:

1. Reuse an existing local component (resolve aliases/equivalent capabilities first) — reuse, don't rebuild.
2. Use an already-registered component from the registry.
3. Install a suitable existing, trusted, supported component → validate → register → connect to the graph → use.
4. Adapt an existing compatible component via a small persistent adapter/wrapper instead of re-creating it.
5. Create only as last resort — then make it permanent immediately: stable id, canonical location, register, add to the capability index + dependency graph, add provenance, use, and allow future reuse.
6. Never reorganise/recreate already-generated components (no Skill X 2 / new / temp variants); extend the existing one. Never create a second competing knowledge source; connect back to the canonical knowledge/ layer. Promote any reusable artifact out of /tmp/scratch into the permanent ecosystem.

Full policy: ../../knowledge/rules/component-lifecycle.md.

Source / provenance

• Source: PART VII §7.1–7.11 (design principles, taxonomy, episodic, semantic, workspace, knowledge graph, broker, coherence, compression, retrieval optimization, long-term persistence).
• Inferred: none beyond Windows path mapping (memory.db, ChromaDB relative persistence path).