One memory. Every AI tool.

Your AI forgets everything between sessions. TraqrDB fixes that — persistent memory that lives in your Postgres, works with any MCP client, and belongs to you. Not Big Tech. Not the labs. You.

Claude CodeCursorCodexAny MCP clientFree forever
npx traqr-memory-mcp
traqrdb — store
$ ▎
How it works

It learns. It remembers. It's yours.

01 — It learns

Your AI stores what matters.

Bug fixes, your preferences, team conventions — categorized, tagged, and entity-linked automatically. Not text dumps. Structured knowledge that gets smarter the more you use it.

memory_storememory_enhancememory_pulse
traqrdb — store
$ memory_store("Vercel hobby tier has 10s timeout...")
Stored [devops/gotcha]
confidence: 0.91
entity: Vercel (linked, 3 prior mentions)
auto-tagged: #deployment #timeout
Your AI will remember this next time.
02 — It remembers

Next session. Different project. Same you.

Your AI knows your stack, your style, your gotchas. No re-explaining. Multi-strategy retrieval finds what matters in milliseconds — across sessions, across projects.

memory_searchmemory_contextmemory_read
traqrdb — search
$ memory_search("why is my API route timing out")
3 results (semantic + BM25 fused)
#1 score: 0.94 [gotcha]
"Vercel hobby tier: 10s serverless timeout."
source: 3 days ago · cited 4x
✓ Applied. Skipped ~2hrs debugging.
03 — It's yours

Your Postgres. Your memories. No SaaS.

One SQL file on your own Postgres + pgvector. No vendor lock-in. No monthly bill. No sending your data to someone else's server. BYO Supabase, RDS, or Docker.

memory_browsememory_auditmemory_archive
terminal — setup
# One file. Any Postgres + pgvector.
$ psql -f setup.sql
✓ 6 tables created
✓ 10 RPCs deployed
✓ pgvector HNSW indexes built
$ SUPABASE_URL=... npx traqr-memory-mcp
TraqrDB ready. 10 tools registered.

10 MCP tools. Zero lock-in.

Works with Claude Code, Cursor, Codex, and any MCP client. Your AI gets memory. You keep control.

memory_store

Store a learning. Auto-derives category, tags, topic.

memory_search

Search by meaning. Semantic + BM25 fused retrieval.

memory_read

Expand a memory. Full content + version history.

memory_context

Assembled context for the current task.

memory_enhance

Deepen existing knowledge with new context.

memory_pulse

Batch capture multiple learnings at once.

memory_browse

Navigate by domain/category. Zero embedding cost.

memory_audit

System health, stats, quality metrics.

memory_archive

Archive stale content. Was correct, now outdated.

memory_forget

Permanently forget incorrect content.

Built on Postgres + pgvectorWorks with any MCP clientOpen source coreBYO database

Give your AI a memory.

Free. Open source. Works with Claude Code, Cursor, Codex, and any MCP client. Your Postgres, your memories, forever.

npx traqr-memory-mcpView on GitHub