ai-chat

Chat with & control AI Agents via telegram, chat UI & voice.

Note: OpenCode is the first supported agent (served over its HTTP/SSE API); the adapter interface keeps other CLI agents open.

Running

• Start the Telegram bot with go run ./cmd/ai-chat start --config ./config.json
• Run the in-process regression suite with go run ./cmd/ai-chat test
• Run one named regression with go run ./cmd/ai-chat test --scenario "Session switching regression"
• Run the explicit Telegram acceptance path with go run ./cmd/ai-chat test --scenario telegram-acceptance --config ./config.json

Deployment (systemd user service)

Run ai-chat as a self-starting systemd user service (validated on ndovu). The binary is expected at ~/.local/bin/ai-chat; the unit resolves ~ to your home directory.

1. Install the binary:

   • Build from source: ./scripts/build.sh && mkdir -p ~/.local/bin && cp ai-chat ~/.local/bin/
   • Or download a release asset from GitHub releases and place it at ~/.local/bin/ai-chat
   • Or npm install -g @peteretelej/ai-chat and copy/link the installed ai-chat binary to ~/.local/bin/ai-chat

2. Place the config at ~/.config/ai-chat/config.json (the service reads this path by default; no --config flag needed). It holds the bot token, so restrict it: chmod 600 ~/.config/ai-chat/config.json

3. Install and start the unit:

   mkdir -p ~/.config/systemd/user
   cp deploy/ai-chat.service ~/.config/systemd/user/
   systemctl --user daemon-reload
   systemctl --user enable --now ai-chat

4. Keep the service running after logout: loginctl enable-linger $USER

5. Watch logs with journalctl --user -u ai-chat -f (daily JSON log files also land in ~/.config/ai-chat/logs)

Restart after a binary or config change with systemctl --user restart ai-chat (./scripts/restart.sh does this for you when the unit is installed).

Continuity: each workspace's opencode serve port is derived deterministically, so opencode attach http://127.0.0.1:<port> URLs stay stable across service restarts; attach again with the same URL after a restart or reboot.

Validation

• Final refactor validation is go test ./...
• Then run ./scripts/pre-push
• Then run one explicit telegram-acceptance pass against a real Telegram bot/chat configuration

Telegram Acceptance

• telegram-acceptance is opt-in and is not part of normal go test or default ai-chat test runs
• By default it uses telegram.bot_token
• You can override the bot with telegram.acceptance_bot_token
• You should set telegram.acceptance_chat_id to the real chat to validate
• If you omit telegram.acceptance_chat_id, the command only falls back when telegram.allowed_users contains exactly one direct-chat user ID

Runtime Notes

• The in-process test harness boots the same store, router, session manager, and outbound response flow used by start
• Telegram remains a transport/rendering adapter on top of router and session-manager behavior