context + torch

Give your agents a memory.

contorch is the open-source memory layer for coding agents — the only one with ears. Meetings transcribe themselves (no bot in the call), notes and lessons accumulate, and any agent that speaks MCP (Claude Code, Codex, anything) picks up where you left off — in any session, on any day.

$ curl -fsSL contorch.com/install | bash

Free · Apache-2.0 · local-first · see how it runs ↓

the context layer

## Where the context comes from

Four streams, all landing on your disk. Two of them need nothing from you at all.

Conversations automatic

Your meetings, word for word.

Take the call like always. contorch hears your mic go live and transcribes both sides — Me and Them — into Markdown. Every promise, decision, and number survives the meeting. No bot joins.

Them  can we ship it before the freeze?
Me    done — it ships tonight.

Docs & links one drop

The spec lives with the task.

Attach files, PRs, dashboards, and threads to the work they belong to — one call, and every future session finds them.

+ add_source auth-spec-v2.pdf → auth-refactor

Decisions & notes one line

Say it once, never again.

Write “JWT ships behind a flag” one time. Stop re-explaining your project's state to every fresh chat window.

+ note "JWT rotation ships behind a flag"

Lessons agents learn automatic

Hard-won knowledge stays won.

When an agent works out your test command or a deploy gotcha, it saves the lesson. The next agent — or the next laptop — doesn't rediscover it.

+ lesson "staging needs the flag service up first"

## What you get back

Ask like you'd ask a colleague who was in the room. Answers come with receipts.

> what did we decide about pricing?

Usage-based; the annual discount is parked until Q4. → meeting-2026-07-08.md, 14:31

> pick up the auth-refactor work

Loads the full manifest — 3 sources, 2 meetings, 1 lesson — and starts where you stopped: the token-rotation PR.

> why does staging keep failing?

A lesson saved on Jun 12: staging deploys need the flag service up first. → repo knowledge, context-orchestrator

Most of what an agent needs already happened —
in a conversation, a doc, a lesson learned.
contorch just keeps it.

## Install

One command stands up the whole stack. Re-run it anytime; it's idempotent.

1 · Bootstrap

$ curl -fsSL contorch.com/install | bash

Installs the context-orchestrator MCP server, the meeting-capture daemon, a menu-bar monitor, and the Claude Code auto-context hook. Missing prerequisites arrive via Homebrew.

2 · Click what macOS asks you to

Two permissions for meeting capture: Screen & System Audio Recording and Microphone. The script ends with a punch-list, and doctor commands verify every piece.

$ transcript-watcher doctor
 all checks passed

3 · Optional: Gemini key

Meeting transcription runs on Gemini — a free key from aistudio.google.com/apikey, about $0.0002 per meeting-minute beyond the free tier. Skip it and everything else stays fully local: tasks, sources, search, clipboard transcripts.

Then restart Claude Code and just work.