Skip to content

What Meshpoint is

Meshpoint is shared memory for coding agents. A connected agent records what each session learns about a repository and loads it into the next session, for anyone on the team, in any harness.

Every agent on your team starts where the last one stopped.

A coding agent forgets everything when its session ends. The decision it made at 3pm, the build gotcha it hit, the convention it worked out from the code: gone. The next session, yours or a teammate’s, rediscovers all of it. Switch harnesses and even your own notes do not follow.

Meshpoint gives a repository one memory that every harness reads and writes.

  1. Session start. The harness asks Meshpoint for the repository’s items on the current branch and for the teammates working on it right now. Both land in the agent’s context before the first prompt.
  2. During the session. Prompts and replies are captured as raw turns, with the file paths the agent touched. Teammates see the agent as live on that branch.
  3. Checkpoint. After twenty replies, the harness asks the agent to write what the session established as items: decisions, gotchas, conventions, facts, status and todos. The agent runs one command and continues.
  4. Next session. Anyone on the team, in any harness, starts with those items.

The platform runs no model. Your agent writes the memory; Meshpoint stores it, ranks it and delivers it.

Harness Integration Status
Claude Code hooks + skill verified
Codex hooks + rules + skill verified
pi extension + skill verified
opencode plugin + skill verified
OpenClaw hook pack verified
Cursor hooks + skill ported from the published hook docs
Antigravity hooks + skill ported from the published hook docs

One command connects each: meshpoint connect <harness>. See Connect a harness.