Claude Code, Codex CLI, and Gemini CLI as your Telegram assistant.
Persistent memory. Scheduled tasks. Live streaming. Docker sandboxing.
Uses only official CLIs. Nothing spoofed, nothing proxied.

Quick start · Why ductor? · Features · Prerequisites · How it works · Commands · Docs · Contributing

---

ductor runs on your machine, uses your existing CLI authentication, and keeps state in plain JSON/Markdown under ~/.ductor/.

You can:

• chat from Telegram with Claude/Codex/Gemini,
• stream responses live into edited Telegram messages,
• run cron jobs and webhooks,
• let heartbeat checks proactively notify you,
• isolate runtime with Docker.

_{Left: /start screen, right: quick action callbacks}

Quick start

pipx install ductor
ductor

The onboarding wizard handles CLI checks, Telegram setup, timezone, optional Docker, and optional background service install.

Why ductor?

ductor executes the real provider CLIs as subprocesses. It does not proxy or spoof provider API calls.

• Official CLIs only (claude, codex, gemini)
• Rule files are plain Markdown (CLAUDE.md, AGENTS.md, GEMINI.md)
• Memory is one Markdown file (workspace/memory_system/MAINMEMORY.md)
• Automation state is JSON (cron_jobs.json, webhooks.json)

Features

Core

• Real-time streaming with live Telegram edits
• Provider/model switching with /model (provider sessions are preserved per chat)
• @model directive support (Claude + Gemini IDs/aliases)
• Media-aware input flow (text/media handling in Telegram layer)
• Inline callback buttons ([button:Label])
• Queue tracking with per-message cancel while lock is held
• Telegram forum topic support (message_thread_id propagation)

Automation

• Cron jobs: in-process scheduler with timezone support, per-job overrides, quiet hours, dependency queue
• Webhooks: wake (inject into active chat) and cron_task (isolated task run) modes
• Heartbeat: proactive checks in active sessions with cooldown + quiet hours
• Cleanup: daily retention cleanup for telegram_files/ and output_to_user/

Infrastructure

• Built-in service manager:
  • Linux: systemd --user
  • macOS: launchd Launch Agent
  • Windows: Task Scheduler task
• Docker sidecar sandbox support (Dockerfile.sandbox)
• Restart protocol (EXIT_RESTART = 42) + sentinel-based user notifications
• Version/update flow with Telegram /upgrade callback path

Developer UX

• Auto-onboarding on first run
• Deep-merge config upgrades (new keys auto-added)
• Rich diagnostics (/diagnose, /status)
• Interactive /cron, /model, /showfiles flows
• Cross-tool skill sync (~/.ductor/workspace/skills, ~/.claude/skills, ~/.codex/skills, ~/.gemini/skills)

Prerequisites

Requirement	Details
Python 3.11+	python3 --version
pipx (recommended) or pip	pip install pipx
At least one CLI installed + authenticated	Claude, Codex, or Gemini
Telegram Bot Token	from @BotFather
Telegram User ID	from @userinfobot
Docker (optional)	recommended for sandboxing

CLI references:

• Claude Code CLI: https://docs.anthropic.com/en/docs/claude-code
• Codex CLI: https://github.com/openai/codex
• Gemini CLI: https://github.com/google-gemini/gemini-cli

Detailed installation: docs/installation.md

Docker management

ductor docker enable
ductor docker disable
ductor docker rebuild

• enable / disable: toggles docker.enabled in ~/.ductor/config/config.json
• rebuild: stops the bot, removes Docker container + image, rebuilds on next start

Run as a background service

ductor service install
ductor service status
ductor service start
ductor service stop
ductor service logs
ductor service uninstall

Platform details:

• Linux: ~/.config/systemd/user/ductor.service, logs via journalctl --user -u ductor -f
• macOS: ~/Library/LaunchAgents/dev.ductor.plist, log files at ~/.ductor/logs/service.log + service.err, ductor service logs tails recent lines from agent.log (fallback newest *.log)
• Windows: Task Scheduler task ductor (10s logon delay, restart-on-failure retries), prefers pythonw.exe -m ductor_bot, ductor service logs tails recent lines from agent.log (fallback newest *.log)

How it works

You (Telegram)
  -> aiogram (AuthMiddleware + SequentialMiddleware)
  -> TelegramBot handlers
  -> Orchestrator (commands/directives/flows)
  -> CLIService
  -> claude/codex/gemini subprocess
  -> streamed or non-streamed response back to Telegram

Background observers run in the same process:

• CronObserver
• HeartbeatObserver
• WebhookObserver
• CleanupObserver
• CodexCacheObserver
• GeminiCacheObserver
• UpdateObserver (only for upgradeable installs: pipx/pip)
• rule sync watcher (CLAUDE.md/AGENTS.md/GEMINI.md)
• skill sync watcher

Session behavior (important):

• sessions are chat-scoped,
• provider sessions are isolated per provider bucket,
• /new resets only the active provider bucket,
• switching back to another provider can resume that provider’s previous session.

Workspace layout

~/.ductor/
  config/config.json
  sessions.json
  cron_jobs.json
  webhooks.json
  CLAUDE.md
  AGENTS.md
  GEMINI.md
  logs/agent.log
  workspace/
    memory_system/MAINMEMORY.md
    cron_tasks/
    skills/
    tools/
      cron_tools/
      webhook_tools/
      telegram_tools/
      user_tools/
    telegram_files/
    output_to_user/

Configuration

Config file: ~/.ductor/config/config.json

Core keys:

Key	Purpose
provider / model	default runtime target
telegram_token, allowed_user_ids	Telegram auth + allowlist
cli_timeout, permission_mode, file_access	runtime execution behavior
reasoning_effort	default Codex reasoning level
gemini_api_key	config fallback key for Gemini API-key mode
docker.*	sandbox settings
heartbeat.*	proactive check settings
cleanup.*	daily cleanup settings
webhooks.*	webhook server settings
cli_parameters.claude/codex/gemini	provider-specific extra args

Full schema: docs/config.md

Telegram bot commands

Command	Description
/start	Welcome screen with quick actions
/new	Reset active provider session for this chat
/stop	Abort active run and drain queued messages
/model	Interactive model/provider selector
/status	Session/provider/auth status
/memory	Show persistent memory file
/cron	Interactive cron management
/showfiles	Browse ~/.ductor/
/diagnose	Runtime diagnostics + cache/log info
/upgrade	Check/apply update flow
/restart	Restart bot
/info	Version + links
/help	Command reference

CLI commands

Command	Description
ductor	Start bot (auto-onboarding if needed)
ductor onboarding	Run setup wizard (smart reset if configured)
ductor reset	Alias for onboarding
ductor status	Runtime status panel
ductor stop	Stop bot + optional Docker container
ductor restart	Stop and re-exec bot
ductor upgrade	Upgrade package and restart (non-dev mode)
ductor uninstall	Remove bot data + uninstall package
ductor service ...	Service management (install/status/start/stop/logs/uninstall)
ductor help	Help + status

Documentation

• docs/README.md
• docs/developer_quickstart.md
• docs/architecture.md
• docs/config.md
• docs/automation.md
• docs/modules/

Disclaimer

ductor runs official provider CLIs and does not impersonate provider clients. Terms and policies can change; validate your own compliance requirements before unattended automation.

Provider policy links:

• Anthropic: https://www.anthropic.com/policies/terms
• OpenAI: https://openai.com/policies/terms-of-use
• Google: https://policies.google.com/terms

Contributing

git clone https://github.com/PleasePrompto/ductor.git
cd ductor
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest
ruff format .
ruff check .
mypy ductor_bot

Target quality bar: zero warnings, zero errors.

License

MIT