Shared context for your coding agents
What one agent learns, every agent knows, before it writes a line.
Your code shows what you built. It doesn't say why.
Every agent starts a session knowing only the code. The reasons behind it, the rules a client set and the work someone is halfway through live in people's heads and old chats, so agents guess.
Code Collie is a shared context layer for AI coding agents. Your agents save those things as you work, and every agent on the project reads them before it writes, whether it runs in Claude Code, Cursor or any other MCP tool.
Say it once. Mention a decision while you work, and your agent saves it to Code Collie with the reason behind it.
Every agent can read it. Your teammate's agent, in a different tool, can read it the moment it's saved.
Works with the AI tools you already use
Add it once and it works in your editor, in a Claude.ai chat and in CI. Your decisions live in Code Collie, not in your repository.
Claude Code
Cursor
Codex
Copilot
CodeRabbit
Any MCP client
Only what the code can't tell you. Anything an agent could work out by reading your code is turned away, so what's kept stays short and worth reading.
Old notes don't linger. If the code stops matching a decision, the decision is flagged. A newer decision about the same thing takes precedence, and the older one is kept.
Decision
One discount code per order.
How it's different
Isn't this what Claude and Cursor memory do?
Their memory belongs to one person, in one tool, on one machine. Code Collie is shared, so a decision saved in Claude Code is there for Cursor, for Codex and for everyone else on the project.
Why not put it in CLAUDE.md or a rules file?
Those files load into every message whether they matter or not, so they grow until agents ignore them, and they only reach tools that read that file. Code Collie hands an agent only what bears on the change it's making.
Why not keep it in Notion or a wiki?
Agents don't read your docs, and docs go stale without anyone noticing. Code Collie is read by agents at the moment they act, and it flags a decision when the code stops matching it.
For people and small teams building with AI agents.
Use it on your own with nothing to configure, or bring your team in when you're ready. It starts empty and fills up as you work.
Be one of the first to use it.
We're letting people in a few at a time so we can work closely with each one.