Shared memory for your org
Instructions, files, and working context in shared collections — so your teammate’s agent already knows what yours figured out on Tuesday. No more re-pasting the same context into six different CLAUDE.md files.
SEC.00 / INDEX — HUTCH.HOMEPAGE()
Vibe
Databasing.
Shared memory for your agents, your team, and every app you work in.
- Add Hutch to your AI clients and agents.
- Tell any of them “save to hutch”.
- Every other app, agent, and session can read it.
BONUSautomagically generated web views — kanban, calendar, timeline, gallery.
100% open source —
audit the code or self-host it.
~/hutch/demo.webm 1600 × 676 · LOOP
SEC.01 / JOBS — WHO.SAVES.WHAT
One database. Three jobs.
People point Hutch at three kinds of work. All of them start with “save this to hutch”.
Markdown is not a database
When an agent pulls in 2,000 records, a .md file is the wrong place for them. One agent collects, another processes, a third reports — same collection, queryable the whole way through.
Speak an app into existence
“Save anything I call a todo to hutch” — and you have a todo app. On your phone, in your browser, in every session. Not a markdown file locked in one repo.
SEC.02 / MEMORY — SHARED.STATE
One memory. Every agent.
Your agents forget everything the moment a session ends. Hutch is the durable layer underneath them all: anything one saves, every other one can read.
WRITE.001 Claude Code terminal · tue WRITE.002 Codex same repo · wed WRITE.003 Claude Desktop your desk · thu
H
READ.001 Any session today, tomorrow, next month
READ.002 Any app CLI, desktop, web, mobile
READ.003 Any agent Claude, Codex, whatever ships next
Across sessions
Context windows close. Collections don’t. Tomorrow’s session picks up the todos, decisions, and open threads today’s session left behind.
Across your apps
Save from Claude Code in the terminal, read it in Claude Desktop at your desk, check it from your phone. Same collections everywhere you work.
Across agents
Claude and Codex working the same backlog — one writes, the other reads. No copy-paste relay between vendors.
SEC.03 / TOOLS — VIEWS.RENDER()
AI-native productivity tools
Memory you can only query is a junk drawer. Everything your agents save renders as a real tool — kanban board, calendar, timeline, gallery — with no schema setup. Say it in prose; work it as a board.
{ task: "ship v2", due: "Fri" }
- Acme: $4.2k MRR
- Globex: trial 6/2
- Initech: churned
marketing sync · tue 2pm · roxie's
▦ Table 01
▤ Kanban 02
▧ Calendar 03
▬ Timeline 04
▩ Gallery 05
~/todos ● ACTIVE
> "save anything i call out as a todo to hutch" A running task list. Table, kanban, or calendar — your call.
→
~/triage ● ACTIVE
> "as we work, save any feature ideas or bugs to hutch" A triaged backlog that survives between coding sessions.
→
~/docs-debt ● ACTIVE
> "as i make changes to the code, save any docs that need updating to hutch" A running checklist of doc gaps you can clear in one sitting.
→
~/support-bugs ● ACTIVE
> "analyze the support tickets and save anything that points to a code issue to hutch" A bug backlog sourced from real customer pain, not guesswork.
→
SEC.04 / CLIENTS — MCP.INTEROP
Use Hutch from
Any MCP-aware client works. Connect once; everything you tell it to save flows through.