You already solved this. Find which thread.
Callimachus quietly indexes every conversation your AI agents write — across eleven tools — into one fast, local, searchable catalogue. Keyword and meaning. No cloud, no account, no tracking.
v0.11.0 · Free · macOS, Windows & Linux · all builds
Eleven agents, one index
- № 001Claude Code
- № 002Codex
- № 003Cursor
- № 004Gemini CLI
- № 005Qwen Code
- № 006Goose
- № 007OpenCode
- № 008Continue
- № 009Cline
- № 010Roo Code
- № 011Kilo Code
- № 001Claude Code
- № 002Codex
- № 003Cursor
- № 004Gemini CLI
- № 005Qwen Code
- № 006Goose
- № 007OpenCode
- № 008Continue
- № 009Cline
- № 010Roo Code
- № 011Kilo Code
Your best debugging session is three tools and two weeks ago. The fix is in there — in a Claude Code thread, or was it Cursor? Callimachus is the index card that finds it in a keystroke, so you stop re-explaining yourself to robots.
Filed, cross-referenced, instant.
One local index behind every surface. Built for the way you actually go looking for something you half-remember.
Find it by what it was about
Keyword search (SQLite FTS5) fused with on-device semantic similarity, so a vague memory — “that vector index migration” — surfaces the right thread even when you've forgotten the exact words. Filter by source, project, recency, starred, or tag — or type file:embed/mod.rs to find every thread that touched a path. Subagent transcripts stay out of the way until you ask for them.
It never leaves your machine
The index, the embeddings, the search — all on disk, on your computer. No account, no sync, no telemetry. The quietest tool in your stack.
Always current, never in the way
A background watcher catalogues new sessions as you work. Reindex any source on demand; forget it's running the rest of the time.
Hand the past to the present
Copy a thread's packed context, insert it into your editor, export it to Obsidian, or let the MCP server feed it to an agent on demand. Your history becomes raw material again, not just a read-only archive.
Distill the lessons, not just the logs
Free heuristic TODO extraction, plus opt-in LLM distillation of decisions, gotchas, and summaries — with cross-thread semantic recall of past decisions and gotchas. Needs local Ollama (keyless) or a cloud API key.
A cited answer over your own threads
Ask a question and get a synthesized, cited answer drawn from your own sessions, with [thread N] citations back to the sources it used. Retrieval-augmented over the local index; needs an LLM engine enabled.
See which commits a conversation produced
Callimachus infers — entirely on-device — which commits a thread shipped, by overlapping the files it discussed with git log inside the thread's time window. Walk the thread-to-commit timeline from the app or cal commits, and ask “which commit came out of this conversation?” straight from an agent.
Resumable cross-tool handoff
Durable checkpoints of a thread — packed transcript plus carry-forward project memory — so work survives a context-window compaction or a jump to another tool. Auto-captured via Claude Code hooks, or snapshot and resume by hand.
Catch contradictions before they ship
Record a decision with the rationale behind it, then let the guard surface settled calls on a topic before an agent quietly re-litigates one. Check a proposal against your own history — from cal check or the MCP — so the past gets a vote before the code does.
Run on your CLI subscription
Distillation, Ask, and the in-app chat can run on your logged-in Claude Code or Codex CLI instead of a cloud API key. Pick it in Settings or the chat provider menu; Callimachus shells out to the CLI you already pay for, nothing new to configure.
The history behind a diff
cal audit-pr returns one JSON bundle for an external review tool: which session produced each commit, prior threads per changed file, and the errors you keep hitting, so a local PR audit sees the context a diff alone hides.
Every thread your agents have written, filed by source and project. Keyword fused with on-device meaning, so the half-remembered one is a keystroke away.
Each repo's decisions, gotchas, and open TODOs, distilled across every session into a durable memory your agents can actually read.
A synthesized, cited answer drawn from your own past sessions, with [thread N] references back to the source it used.
A year of work at a glance, and the mistakes you keep re-running across tools, so you fix the pattern instead of the instance.
Threads, messages, and semantic coverage at a glance, plus an estimate of what your AI coding actually cost, broken down by model and by your priciest threads.
A provider-agnostic chat that searches your own history before it answers. Bring a key, or run it keyless on your logged-in Claude Code or Codex CLI.
Four ways into the same catalogue.
One index, read from wherever you already are.
- № 01Desktop app
A native app for macOS, Windows, and Linux. Browse, search, and read every thread; chat over your own history; export to Obsidian. Auto-updates.
- № 02VS Code & Cursor
A sidebar and transcript tabs inside your editor. Same local index, no context-switch. Works in VS Code, Cursor, and VSCodium.
- № 03cal CLI
`cal search`, `cal recent`, `cal cat` — grep your past sessions from the terminal and pipe context straight into the next agent.
- № 04MCP server
Expose your indexed history to any MCP client. Agents can search your past work and pull the thread they need, on demand.
A library that doesn't read over your shoulder.
Cookieless, aggregate analytics on this site — no tracking cookies, no cross-site profiles. No telemetry in the app. Your conversations are indexed where they already live — on your machine — and stay there. Open source under AGPL-3.0, so you never have to take our word for it.
Questions, briefly answered.
Open the catalogue.
Free, local, and quietly yours. Run the app once and every thread you've ever written is a keystroke away.
v0.11.0 · Free · macOS, Windows & Linux · all builds