TagOpen

Self-hostable, channel-native AI teammate for Slack.
LLM-agnostic open source alternative to Claude Tag.

One agent per channel, shared by the whole team. Persistent memory. Skill auto-creation. Ambient monitoring. No vendor lock-in.

What makes this different

Most Slack AI bots are personal assistants — one context per user, isolated DMs. TagOpen flips this:

• Channel-scoped identity — one agent shared by everyone in #engineering. All users see the same context, pick up mid-thread.
• Multi-user attribution — every message is tagged [@alice] so the agent knows who said what and can follow up with the right person.
• Agent-curated memory — after each conversation, the agent decides what to persist to MEMORY.md. No noisy append-only logs.
• Skill auto-creation — after complex multi-step tasks, the agent writes a SKILL.md capturing what it learned. Institutional knowledge accumulates automatically.
• Ambient heartbeat — configurable proactive monitoring: the agent surfaces stale threads, approaching deadlines, and unresolved questions without being tagged.
• File-based config — each channel is a directory of Markdown files. Version-controllable, no UI required.
• MCP-native tools — plug in any MCP server per channel. Admins control exactly what each channel's agent can access.

Quickstart

# 1. Clone and install
git clone https://github.com/Anil-matcha/tagopen
cd tagopen
pip install -e ".[dev]"

# 2. Configure
cp .env.example .env
# Fill in SLACK_BOT_TOKEN, SLACK_APP_TOKEN, ANTHROPIC_API_KEY

# 3. Set up your first channel config
mkdir -p data/channels/YOUR_CHANNEL_ID
cp channels/example/CHANNEL.md data/channels/YOUR_CHANNEL_ID/CHANNEL.md
# Edit CHANNEL.md to describe your channel's purpose

# 4. Run
tagopen

Then @tagopen in your channel.

Channel configuration

Each channel gets a directory under data/channels/<channel_id>/:

data/channels/C01234ABC/
  CHANNEL.md      ← identity, purpose, tone
  MEMORY.md       ← agent-maintained facts (do not edit manually)
  tools.toml      ← which MCP servers are enabled
  skills/         ← auto-created skill playbooks
    deploy.md
    oncall.md

CHANNEL.md example

# Engineering Channel

You are the engineering team's AI teammate.
Be concise, technical, and ask before triggering deploys.

## Team context
- Stack: Python, React, PostgreSQL, AWS
- We do not deploy on Fridays

tools.toml example

[[mcp_server]]
name = "github"
url = "mcp://localhost:3001"
allowed_tools = ["list_prs", "get_file", "create_comment"]

Architecture

Slack (Socket Mode)
       ↓
  Bolt Gateway
       ↓
  Channel Router  ← (workspace_id, channel_id) → AgentSession
       ↓
  Context Assembler  ← CHANNEL.md + MEMORY.md + skills + recent msgs
       ↓
  Agent Loop (ReAct + tool-use via LiteLLM)
       ├── Tool Registry (built-ins + MCP)
       ├── Streaming reply → Slack thread
       ├── Memory curation turn (Letta inner loop)
       └── Skill auto-creation (Hermes pattern)
       ↓
  SQLite + FTS5  ← per-channel message store
       ↓
  Ambient Engine  ← heartbeat cron, proactive posts

Supported LLMs

Uses LiteLLM — swap provider via LLM_MODEL in .env:

Per-channel override — different channels can use different models. Add to data/channels/<id>/tools.toml:

[llm]
model = "gpt-4o"

Development

# Run tests
pytest

# Lint
ruff check .

# Type check
mypy tagopen/

Roadmap

• Phase 1 — Channel-native reactive teammate
• Phase 2 — Mem0 semantic recall + skill curator
• Phase 3 — Ambient heartbeat + agent-managed crons
• Phase 4 — Admin web UI + token governance
• Phase 5 — Discord + Teams adapters

See PLAN.md for full architecture and design decisions.

License

MIT