Give your AI agent a voice on every chat platform.
A lightweight daemon that lets AI agents send, receive, and stream messages across Slack, Discord, Mattermost, Telegram, WhatsApp, IRC, Matrix, Twilio, and Zulip through a single interface.

Website · About · Quick Start · Platform Setup

AI agents need to communicate with humans where they already are - Slack, Discord, Mattermost, Telegram, WhatsApp, IRC, Matrix, Twilio, Zulip. But every platform speaks a different protocol. Building an agent that can participate in conversations across all of them means writing and maintaining separate integrations before your agent can even say "hello."

Pantalk gives your AI agent a single, consistent interface to all chat platforms. One daemon (pantalkd) handles the upstream complexity - auth, sessions, reconnects, rate limits - while your agent talks through simple CLI commands or a Unix domain socket with a JSON protocol.

graph TD
    Agent["Your AI Agent<br/><em>(any language, any framework)</em>"]
    Agent -->|send| Socket
    Agent -->|history| Socket
    Agent -->|notify| Socket
    Agent -->|stream| Socket
    Socket["Unix Domain Socket<br/><em>(JSON protocol)</em>"]
    Socket --> Daemon["pantalkd<br/><em>(daemon)</em>"]
    Daemon --> Slack
    Daemon --> Discord
    Daemon --> Mattermost
    Daemon --> Telegram
    Daemon --> WhatsApp
    Daemon --> IRC
    Daemon --> Matrix
    Daemon --> Twilio
    Daemon --> Zulip
    Daemon --> More["..."]

All clients connect to pantalkd through a Unix domain socket using a simple JSON protocol. AI agents and LLM tools can send, receive, and stream chat messages without embedding any service SDK.

• Agent-first - structured output, skill definitions, and notification routing designed for AI agents
• One daemon, all platforms - upstream auth/session complexity lives in pantalkd
• Composable CLI - JSON over Unix socket, works with grep, jq, xargs, and any language
• Multi-bot - define multiple bots per service via config
• Local-first - SQLite persistence, no external dependencies

cmd/
  pantalkd/              # Daemon entry point
  pantalk/               # Unified CLI (messaging + admin)
configs/
  pantalk.example.yaml   # Example configuration
docs/
  agents.md              # Reactive agent configuration guide
  slack-setup.md         # Slack platform setup guide
  discord-setup.md       # Discord platform setup guide
  mattermost-setup.md    # Mattermost platform setup guide
  telegram-setup.md      # Telegram platform setup guide
  whatsapp-setup.md      # WhatsApp platform setup guide
  irc-setup.md           # IRC platform setup guide
  matrix-setup.md        # Matrix platform setup guide
  twilio-setup.md        # Twilio platform setup guide
  zulip-setup.md         # Zulip platform setup guide
  claude-code-hooks.md   # Claude Code hooks integration guide
internal/
  client/                # Shared IPC client logic
  config/                # YAML parsing & validation
  protocol/              # JSON protocol types
  server/                # Daemon server + SQLite
  upstream/              # Platform connectors

Create a config file with your bot credentials:

mkdir -p ~/.config/pantalk
cat > ~/.config/pantalk/config.yaml << 'EOF'
server:
  notification_history_size: 1000

bots:
  - name: my-bot
    type: slack
    bot_token: $SLACK_BOT_TOKEN
    app_level_token: $SLACK_APP_LEVEL_TOKEN
    channels:
      - C0123456789
EOF

See configs/pantalk.example.yaml for a full example with all platforms.

# Uses ~/.config/pantalk/config.yaml by default
pantalkd &

# Or specify a custom config
pantalkd --config /path/to/pantalk.yaml

# Override socket/db paths
pantalkd --socket /tmp/pantalk-dev.sock --db /tmp/pantalk-dev.db

All paths follow the XDG Base Directory Specification.

The unified pantalk binary works with all platforms. The daemon resolves which service each bot belongs to automatically.

# List all bots across all services
pantalk bots

# Send a message (service is auto-resolved from bot name)
pantalk send --bot my-bot --channel C0123456789 --text "hello from cli"
pantalk send --bot my-bot --channel C0123456789 --thread 1711234567.000100 --text "reply in thread"

# Read history
pantalk history --bot my-bot --channel C0123456789 --limit 20

# Check & clear notifications
pantalk notifications --bot my-bot --unseen --limit 50
pantalk notifications --bot my-bot --unseen --clear

# Stream events in real-time (auto-disconnects after 60s by default)
pantalk stream --bot my-bot --notify

# Stream with custom timeout (0 = no timeout)
pantalk stream --bot my-bot --notify --timeout 120

Tip: JSON output is automatic when stdout is not a terminal (e.g. when called by an AI agent). Use --json to force it in interactive mode.

# Validate config (uses default config location)
pantalk validate

# List configured bots without exposing credentials
pantalk config list-bots --json

# Edit non-interactively
pantalk config set-server --history 1000

pantalk config add-bot \
  --type slack --name my-bot \
  --bot-token '$SLACK_BOT_TOKEN' --app-level-token '$SLACK_APP_LEVEL_TOKEN'

# Hot-reload running daemon
pantalk reload

pantalkd initializes entirely from YAML config with strict schema validation:

• ❌ Unknown keys → config load failure
• ❌ Missing required provider fields → fast failure
• ✅ transport and endpoint optional for built-in providers (Slack, Discord, Telegram)
• ⚠️ Mattermost requires endpoint on the bot entry

bots:
  - name: ops-bot # --bot ops-bot
    type: slack
  - name: eng-bot # --bot eng-bot
    type: slack

pantalk reload

• Reloads config from the daemon's --config path
• Restarts service connectors in-process
• Supports bot/service changes
• Does not switch socket_path or db_path at runtime (restart pantalkd for those)

JSON over Unix domain socket. Every request is a single JSON object with an action field:

{"action": "bots"}
{"action": "send", "bot": "my-bot", "channel": "C0123", "text": "hello"}
{"action": "history", "bot": "my-bot", "channel": "C0123", "limit": 20}
{"action": "history", "bot": "my-bot", "search": "deploy", "limit": 50}
{"action": "notifications", "bot": "my-bot", "unseen": true}
{"action": "subscribe", "bot": "my-bot", "notify": true}

All events are persisted locally in SQLite. history always reads from local state.

Pantalk surfaces events relevant to the agent via notifications. This is designed for AI agents that need to know when they're being talked to.

notifications --bot my-bot --clear                       # All for a bot
notifications --bot my-bot --channel C0 --clear          # Scoped by channel
notifications --clear --all                              # Everything
history --bot my-bot --clear                             # Clear history for a bot
history --clear --all                                    # Clear all history

An inbound event becomes a notification when any of these are true:

• Direct message - target matches dm:*, direct:*, user:*, or DM-like channel IDs
• Mention - message contains @bot-name or <@platform-user-id> (auto-discovered at runtime)
• Active thread - event is on a route where the agent previously sent a message

Each platform requires its own app/bot setup before Pantalk can connect. See the detailed guides:

• Richer provider event support (edits, reactions, thread metadata)
• Provider-specific message normalization
• Additional platform connectors

MCPShim - Use any MCP server as a standard CLI command. Pantalk gives your agent a voice; MCPShim gives it tools. Together they form a complete agent infrastructure stack.

pantalk.dev