call me vera — shared memory for AI coding agents
how it fits together
Claude ──┐ Local ────┼── same .vera_store.db (SQLite, WAL) ── numbered, append-only memory Codex ────┘ author: claude | local | vera
🙈 no interpretation
No fact extraction, no contradiction detection. Deciding what an entry means is the reading agent's job, every time — not a fixed rule baked into vera.
🔢 citable by number
Every REQUEST / CHANGE / REASON / RESULT / STATE becomes its own permanent, numbered row. Cite `#12`, fetch it exactly with `vera lookup`.
🔒 append-only, for real
SQLite triggers reject UPDATE/DELETE on the log — not a convention, a database-level guarantee.
🏷️ named, resumable
Name a memory once; any other session or app reconnects to the exact same store by name, regardless of its own default path.
quick start
pip install -e ".[mcp]" vera init . claude mcp add vera -- vera mcp --store ./.vera_store.db Connect vera to your agent, then say:
Vera guide vera explains itself directly to the agent — in whichever of 12 languages you asked in — including the exact behavior protocol to follow and when to suggest compression.
commands
vera start call before work — memory + agent protocol vera record the structured "vera" capture, auto-dedups a retry vera lookup <n>fetch one entry by its citation number vera search text search, results carry citation numbers vera compress store an AI-authored digest citing entry numbers vera claim-name name this memory so any session/app can resume it vera stats entries, size vs. compression threshold, state vera mcp start the MCP server