amplifthq/opentag

What it solves

OpenTag provides a way to integrate an AI coding agent into a Slack channel as a persistent "teammate." It solves the problem of agent dashboards being disconnected from team communication and the risk of giving cloud-based agents full access to local development environments. By using a self-hosted control plane, it ensures that the AI agent remains reachable in Slack even when the actual computer running the code (the Runner) is offline.

How it works

OpenTag uses a three-part architecture: a self-hosted Control Plane, a paired Runner, and a coding agent (via ACP).

1. Slack Integration: Users mention the agent in Slack to queue work.
2. Control Plane: Acts as a relay and coordinator, storing the state of work and requests in a PostgreSQL database.
3. Runner: A user-controlled computer that pairs with the Control Plane. When the Runner is online, it pulls the queued work and invokes a coding agent to act on a local checkout of the source code.
4. Execution: The agent performs the work and reports status, decisions, and evidence (such as GitHub draft PRs) back to the Slack thread.

Who it’s for

Engineering teams who want an AI coding assistant that lives in their existing communication channels and operates on local, secure development machines rather than in a cloud-hosted environment.

Highlights

• Persistent Presence: The agent stays in Slack and can report its offline status instead of simply disappearing.
• Local Execution: Source code, credentials, and worktrees remain on the user's own Runner machine.
• Fact-Based Reporting: Reports actual executor output and GitHub publications rather than simulated success.
• Governed Access: Writes to the codebase are handled through explicit authority, such as draft PRs, preventing auto-merging.
• Agent-Friendly Setup: Includes an agent-readable install guide and specific skills to help other AI agents (like Claude Code) assist with the local setup.