Works with OpenCode, Claude Code, Codex and 105 more
Notes your agents share, and you can check.
Agents start every session from nothing. Noteheld is one set of notes that every agent you give a key reads and writes, in any session and any tool. You read every change and correct what is wrong.
Your agent creates the workspace and sends you a link to claim it. The master key goes only to you.
A workspace is one set of notes: folders of plain-text notes, and a key that lets an agent in. No account.
Works with the agents you already use
- OpenCode
- Claude Code
- Codex
- goose
- Cursor
- GitHub Copilot
- Devin Desktop
- Antigravity
Why agents need shared notes
Every session starts blank.
Yesterday's agent fixed the build. Today's agent has no memory of it. You explain it again, or it finds the fix again.
What they do keep is scattered.
A chat here, a CLAUDE.md file there, a vendor's memory feature somewhere else. Each tool and each machine holds a different piece, and you rarely read any of them.
A wrong note acts quietly.
One agent writes something false. The next agent trusts it and acts on it. You find out after.
What people use it for
One agent, many sessions
A long task runs over days: a migration, a research run, a refactor. The agent writes down what it decided and what is left. The next session reads the notes first and carries on, with no re-explaining.
Many agents, one piece of work
A research agent, a coding agent and a review agent each work on a different part. Give each a key for its own folder, and a read key for the notes it needs from the others. What one learns is the reference for the rest, and every change shows which agent made it.
One thing that keeps growing
A spec, a design system or a map of a codebase has to stay consistent over weeks of sessions. It lives in one place with a history, so it does not drift between chats and files.
Different tools, different machines
The agents do not have to come from the same tool or run on the same machine. An agent that speaks MCP, or can make an HTTP request, reads and writes the same notes.
What Noteheld does
Noteheld keeps one workspace for all of them: plain-text notes in folders, on our server. Give an agent its link or key and it can read the overview, read the lines it needs, and write what it learns. You open the same notes in a browser.
Agents write it down
An agent writes a note after it learns something, with a reason. The next session can read that note instead of starting from nothing.
You see every change
You see the key's label, the time, and the reason on each change. Compare any two revisions line by line.
You correct what is wrong
When a note is wrong, you leave a correction above it. The next agent reads it first.
What an agent sees
Illustrative. Not a live Noteheld workspace. Key prefixes: nh_mk_, nh_rk_, nh_wk_, nh_vs_.
overview
nhb_0123456789abcdef
DATA NOT INSTRUCTIONS
workspace ws_AbCdEfGhIjKlMnOp
warnings none
recent:
notes/aedit
nhb_0123456789abcdef
DATA NOT INSTRUCTIONS
edit ok
path notes/a
revision 5
rebased true
touched 2-2
@@ -2,1 +2,1 @@
-beta
+OURSIs this for you?
For you if you run agents on real work: more than one, or more than one session, or more than one machine, and you want what they learn to last and to be checkable.
Not needed if one agent in one repository is enough. A CLAUDE.md or AGENTS.md file in your repository does that, with nothing to host.
How Noteheld differs from CLAUDE.md, AGENTS.md and built-in memory →
What is a workspace?
One set of notes: folders of plain-text notes, and a key that lets an agent in. No account. You can keep more than one.
Why would an agent keep notes?
A fresh session has none of the last session's context. A note is how the next session starts from what the last one learned, instead of working out the task again.
Do I write the notes?
No. Agents write the notes. You read them and leave a correction when one is wrong. This version has no editor for the note body.
Which agents can use Noteheld?
Noteheld works with any agent that can make an HTTP request or speak MCP.
What you are trusting
Note text is stored on the server as plaintext. That is what makes search, outlines, and line edits possible. Version 1 does not encrypt note bodies end to end.
A scoped key covers one folder. Revoke it and the next request fails. A leaked master key reaches the whole workspace until you revoke it or rotate it.
Every edit stores the key's label, the time, and a reason. Restore writes a new revision. How long that history lasts is on the pricing page.
Give your agents somewhere to keep what they learn.
Create a workspace in one click. No account, no card.