You set the gate
Manual, Hybrid or Auto. Whatever you pick, a rule that contradicts a live one waits for a person.
Shared team memory for AI coding agents.
Harbor gives Claude Code, Cursor, Codex and GitHub Copilot one memory of what your team decided. It learns the rules from Slack, pull request reviews and docs, holds what needs a person behind a gate you set, and hands every session the rules that apply to its task, on every prompt.
Every team starts with a 20-minute setup call with us.
Every agent
One memory, four agents.
One setup per machine wires every agent it finds with hooks and an MCP server. The same rules answer questions in Slack and on the harbor CLI.
| When the rules arrive | How it is wired | |
|---|---|---|
| Claude Code | When the session starts and on every prompt | Hooks and the MCP server. |
| Cursor | When the session starts and on every prompt | Hooks and the MCP server, in a current Cursor app. |
| Codex | When the session starts and on every prompt | Hooks and the MCP server. |
| GitHub Copilot | On every prompt and after each file it touches | Copilot CLI and VS Code agent mode share one hook file. The cloud agent is not supported yet. |
Every agent, file and hook: the agents docs.
Why teams pick Harbor
Decide what gets in. See what gets used.
A memory every agent reads is only as good as what gets into it, and only worth keeping if the agents use it.
You see what got used
Every rule shows how often it was served to a session and how often an answer cited it.
You block a risky command
Block rules refuse it before it runs, in Claude Code, Cursor, Codex and Copilot.
Side by side
Memory for coding agents, compared.
| Learns from | Before agents get it | Seeing what was used | |
|---|---|---|---|
| Harbor | Slack, pull request reviews, docs, and what agent sessions record | A gate you set: Manual, Hybrid or Auto. A contradiction always waits for a person. | Every rule shows how often it was served and how often an answer cited it |
| Supermemory | What its plugins capture from agent sessions | Facts it inferred wait for approval; facts that were stated go in | Not stated on the pages we read |
| Cognee | Agent sessions; Cognee Cloud adds Slack, GitHub and Google Drive | Automatic; no review step documented | Citations, and a coverage score per question on Cognee Cloud |
| Mem0 | What its plugin captures from agent sessions | Automatic; no review step documented | Access logs of what was added, edited or served |
Each product's row is from its own pages, read on 6 October 2026. More in Harbor compared and the best company brain tools for coding agents. Out of date? Tell us.
Questions
Before you wire it up.
What is shared memory for AI coding agents?
One store of what a team has decided that every coding agent reads. A rule agreed once, in a Slack thread or a review, reaches every Claude Code, Cursor, Codex and Copilot session, without anyone copying it into each tool's rules file.
Does Harbor work with Claude Code, Cursor, Codex and GitHub Copilot?
Yes. One setup on each machine wires every agent it finds with hooks and an MCP server. Claude Code, Cursor and Codex get the rules that apply when a session starts and on every prompt. Copilot gets them on every prompt and after each file it touches; Copilot's cloud agent is not supported yet.
How is this different from CLAUDE.md or AGENTS.md?
A rules file is per repository and per tool, loads whole, and someone has to keep it current. Harbor learns rules from where the team decides, serves the ones that apply to the task, and counts how often each one was served and cited. Your rules files keep working beside it.
Who decides what the agents are told?
You do, with the gate. Manual holds everything for a person, Hybrid, the default, adds routine rules and holds the rest, and Auto adds the rest too. A rule that contradicts a live one always waits for a person.
How do we start, and what does it cost?
Every team starts with a 20-minute setup call with us. Free costs nothing for your whole team on one source. Pro is $24 per person a month billed yearly.
Decide once.
Every agent knows.
Nothing to migrate and nothing to write. Every agent starts with what your team settles, and two weeks in you see which rules they cite. Leave an address and we will set your team up on a 20-minute call.