OpenTag

linxidnju/OpenTag is ranked 18,261 among these 8 peers; stanford-scs/jai is ranked 18,268.

OpenTag Open-source, channel-native multiplayer AI gateway with shared, auditable team knowledge. Tag local AI agents into Slack threads, with local execution, visible progress, approvals, audit trails, and pluggable agent runtimes. Instead of running an agent from one person's terminal and pasting results back into chat, a team can mention , discuss the task in the same thread, approve risky actions when needed, and keep the outcome visible to everyone who has access to that channel. Let teammates add context

before or during the agent run. Route work to different agent backends such as Codex, Claude Code, OpenCode, Docker-based agents, HTTP agents, or custom CLIs. Teach OpenTag an explicit team fact, decision, convention, preference, or procedure; corrections are versioned, scoped, sourced, and reused by every configured runtime. Use Channel Profiles / Access Bundles so different projects can have different permissions, memories, default agents, and run limits. Require approvals before write actions or risky operations. Keep a record of sessions, decisions

, outputs, and generated artifacts. OpenTag is designed for team workflows where context, permission, and accountability matter. OpenTag is built around shared channels. **Shared by default**: the conversation, context, and result stay in the team thread. **Channel-aware**: each channel can have its own project, instructions, memory, permissions, default runtime, and run limits. **Shared team knowledge**: correct a fact once and future threads receive the latest active version, with provenance and revision history under your control. **Agent-flexible**: use the agent

runtime that fits the task instead of locking the team into one backend. **Approval-ready**: sensitive actions can pause for human review. **Auditable**: sessions, messages, approvals, outputs, and artifacts can be reviewed later. Try OpenTag locally without Slack: Start the Slack gateway with the example configuration: starts the local daemon when needed and opens a step-by-step setup wizard at . The wizard walks through agent selection, Slack manifest import, local token saving, and final verification. Local console mode for

trying OpenTag without Slack. Runtime options for mock runs, Codex, Claude Code, OpenCode, Docker, HTTP agents, and generic CLI agents. Channel configuration for default runtime, allowed runtimes, allowed users, approvers, instructions, memory, and workspace roots. Versioned Team Knowledge with workspace/private-channel boundaries, source references, archive/restore, and permanent deletion. Channel Profiles / Access Bundles for per-channel runtime, memory, approval, and limit boundaries. Local storage for sessions, messages, approvals, runs, audit records, and artifacts. Admin and integration surfaces for teams

that want to build deeper workflows. - generic CLI, HTTP, and MCP-capable runtime templates For implementation details, see . Secrets should be supplied through environment variables, not committed to the repository. Channels can limit who can use OpenTag and which runtimes are allowed. Write-capable or risky actions can require approval. Workspaces and filesystem roots can be restricted by configuration. Runs, approvals, artifacts, and audit records are kept for review. See for security reporting and project security guidance. It is suitable

for local trials, internal team experiments, and runtime integration work. Planned improvements include Slack OAuth installation, stronger multi-instance storage