charliee1w/consolidation-memory
charliee1w
Python5updated 4mo ago
Python5updated 4mo agoLocal-first persistent memory for AI agents — store, recall, and consolidate knowledge across sessions using FAISS, SQLite, and any LLM. MCP server for coding agents and any MCP client.
ClaudeCodexCursorMCP
agentic-aiai-agentsclaude-codeclaude-desktopcoding-agentscursordrift-detectionfaissllmlocal-firstlocal-llmmcp
README
consolidation-memory
Trust-calibrated working memory for coding agents.
Store what happened. Consolidate what you learned. Recall with provenance — and lower trust when the repo moves on.
consolidation-memory is local-first agent memory built for long-horizon coding work: SQLite + FAISS + markdown topics on disk, one semantic contract across Python, MCP, REST, and OpenAI-style tools, and explicit trust signals instead of opaque similarity search.
At a glance
| | Typical agent memory | consolidation-memory |
| --- | --- | --- |
| Unit of reuse | Embedded chat snippets | Claims — hash-stable beliefs with provenance |
| Raw input | Same blob as retrieval | Episodes — evidence kept separate from beliefs |
| Staleness | Hidden until wrong | Drift challenge when anchored files change |
| Time | Present-tense only | as_of queries for prior belief states |
| Sharing | Hope the filter works | Scope envelopes + persisted policy/ACL primitives |
| Consolidation | Always calls an LLM | Fast path for structured episodes; LLM only for residue |
Focus: drift-aware debugging memory — preserve fixes, recall with trust signals, degrade gracefully after refactors.
Mental model
claims are the reusable unit; episodes are the raw evidence behind them.
Episode Evidence from a