Why coop?

For engineering leads evaluating collaboration tools for AI-assisted development.

The problem

AI coding tools have gotten good at the single-developer loop: you prompt, the AI responds, you iterate. The problem is that most real engineering work isn't a single-developer loop.

When teams adopt AI coding tools, they run into the same wall: every session starts from scratch. There's no shared context, no shared history, no way for one session to know what another found out five minutes ago. Knowledge stays trapped in individual terminals.

This is already happening on teams today:

• A frontend dev using Cursor and a backend dev using Claude Code are working on the same feature with no shared context layer
• A team standardized on Claude Code, but one engineer prefers OpenCode — they're locked out of any collaboration tool that's Claude-specific
• A CI pipeline runs Codex for automated fixes, but can't access the human team's session history
• An architect reviews code in a browser while engineers work in terminal tools — there's no common interface

Existing solutions don't fit

Slack / Teams

Slack is where humans coordinate. It's not structured for code context, there's no hook integration, and AI tools can't read or write to it programmatically without custom middleware. You end up copy-pasting findings into Slack and copy-pasting them back into your terminal. That's exactly the manual overhead AI coding tools are supposed to eliminate.

Git

Git is async only. It captures code, not intent or findings. There's no real-time coordination, no context injection, and committing a "work in progress" message every time you want to share a finding with a teammate is not a workflow anyone wants.

Shared docs (Notion, Confluence, etc.)

Docs go stale immediately. They require manual updates. They're not queryable from a terminal. An AI coding tool has no way to pull from them automatically. They work for architecture decisions that have settled — not for the fast-moving context of an active coding session.

Shared Claude projects

Locked to one LLM provider. No hook integration. No CLI. No cross-tool bridging. Useful for organizing files with one AI, not for coordinating multiple developers using different tools.

What coop does differently

1. Real-time messaging between AI coding sessions. Messages arrive at the next prompt — not in a separate app, not via copy-paste. The AI reads them and acts on them without you doing anything extra.
2. Automatic context injection. A hook fires before every prompt and injects waiting messages as context. The AI sees what your teammates found. You don't have to relay it.
3. File sharing. Screenshots, logs, patches, configs — files flow through the session. No shared drives, no email attachments, no "can you send me that file."
4. Tool-agnostic. Claude Code, Codex CLI, OpenCode, Gemini CLI, Goose — coop sits above the tool layer. It doesn't care which LLM you use or which CLI wraps it. Teams don't need to agree on a tool. They just need to agree on a session.
5. Gateway bridge. OpenClaw integration routes messages to Slack, Discord, Telegram, and 24+ other channels. Non-CLI collaborators (PMs, designers, stakeholders) can follow along without running a terminal.
6. Searchable history. AutoRAG indexes every message and file for semantic retrieval. Run coop session context --query "what was the auth decision?" and get the relevant history back — from the CLI, from the web console, or injected automatically before your next prompt.

What coop is not

Being clear about the boundaries matters. Coop does one thing well and deliberately avoids scope creep.

• Not a replacement for git. Coop is for real-time coordination — passing findings, plans, and context between active sessions. Git handles code history and async review. They complement each other; coop doesn't replace either.
• Not a chat app. Messages are structured context that AI tools can read and act on. There's no emoji reactions, no threads, no read receipts in the social sense. The unit of value is context injection, not conversation.
• Not tied to one LLM or one tool. The whole point is that your team can use different tools and different models and still share a collaboration layer. Coop has no opinion on which AI you use.
• Not a project management tool. Coop doesn't track tasks, sprints, or assignments. It handles the session-level coordination that happens between tickets — the fast, ephemeral context that Linear and Jira aren't designed for.

Who it's for

Coop is useful any time more than one AI coding session is working on the same problem. That's most common in:

• Teams with two or more engineers using AI coding tools on the same feature or service
• Solo developers working across multiple machines or sessions (desktop + laptop, local + remote)
• Teams that mix AI coding tools — one person uses Claude Code, another uses Codex, a third runs locally with OpenCode
• CI/CD pipelines that use AI for automated fixes and need to share findings with human sessions