Journal
What we learn while building shared memory.
Notes on what coding agents remember, where continuity breaks, how grounded recall behaves, and why the human reading surface matters.
How to audit AI coding agent memory before it changes your code
Inspect coding-agent memory sources, retrieval evidence, and delivery state before stale or conflicting knowledge reaches Claude Code or Codex.
Claude Code and Codex shared memory: one local Context across both agents
Give Claude Code and Codex access to the same local, source-attributed project knowledge without copying raw transcripts or moving into another coding agent.
Local-first memory for coding agents, optionally synced
Docmancer keeps your coding agents' memory local by default. Optional cloud sync uploads ciphertext only, so the service never sees the memory itself.
Docmancer 0.7.0: a file-first terminal explorer for agent memory
Browse complete memory and instruction files, inspect documentation, search passages in context, and audit the local corpus from one terminal interface.
Docmancer 0.6.14: the complete local memory loop
Docmancer can now recall, write, inspect, forget, capture, and share coding-agent memory without a hosted server.
Docmancer 0.6.15: recall you can measure and trust
Docmancer now rejects weak memory matches, gates retrieval against a real-workflow corpus, previews capture, and audits corpus health locally.
Local docs RAG on the same engine that powers your agent memory
Index library docs, GitHub repos, and internal Markdown into the same local hybrid retrieval engine Docmancer uses for agent memory. No API key needed.
Claude Code memory search: find every decision your coding agents wrote down
Search Claude Code memory, Codex notes, Cursor rules, and repository instructions from one local index with Docmancer.
Your coding agents have been writing things down. You can't read any of it.
Claude Code, Codex, Cursor, and Windsurf each keep their own memory. Docmancer indexes all of it locally and combines it into reviewed context packs.
Context engineering for coding agents: what the model can see, not what you ask
A concrete framework for context engineering in coding agents, covering four context surfaces, the token cost of each, and how to measure retrieval quality.