A Telegram bridge for terminal-based AI coding agents (Claude, Codex, Gemini) running in tmux or herdr, allowing users to monitor output, send commands, and manage multiple parallel sessions from their phone while maintaining full terminal access. Designed for developers who want to walk away from their machine mid-session and resume work on their phone without losing context or losing control of the underlying terminal.
Telegram ↔ tmux/herdr bridge for Claude Code, Codex CLI, and Gemini CLI. Monitor output, respond to prompts, manage parallel sessions. Control AI coding agents from your phone.
At a glance
Control and monitor AI coding agents running in tmux or herdr from Telegram, enabling phone-based oversight mid-session while preserving terminal access.
uv tool install ccgram
README
CCGram — Control AI Coding Agents from Telegram
Control AI coding agents from your phone. Walk away mid-session. Keep monitoring and responding from Telegram—without losing terminal access.
Why CCGram?
AI coding agents run in your terminal. Other Telegram bots wrap agent SDKs into isolated API sessions you can't resume in your terminal. CCGram is different. It sits on top of your terminal multiplexer (tmux or herdr), not any agent SDK. Your agent process stays exactly where it is—your session is the source of truth.
This means:
- Desktop to phone, mid-conversation — walk away and keep monitoring from Telegram
- Phone back to desktop, anytime — attach to your terminal and you're back with full scrollback
- Multiple sessions in parallel — each Telegram topic maps to a separate tmux window or guarded Herdr agent session
How It Works
graph LR
subgraph phone["📱 Telegram Group (Forum Topics)"]
direction TB
T1["💬 api — Claude"]
T2["💬 ui — Codex"]
T3["💬 data — Gemini"]
T4["💬 ops — Shell"]
T5["💬 lab — Pi"]
end
subgraph bridge["⚡ CCGram"]
direction TB
B1["read output\n(transcripts + terminal)"]
B2["send keystrokes\n(tmux / herdr)"]
B3["instant notifications\n(Claude hooks)"]
end
subgraph machine["🖥️ Your Machine — tmux / herdr"]
direction TB
W1["window @0 · claude"]
W2["window @1 · codex"]
W3["window @2 · gemini"]
W4["window @3 · bash"]
W5["window @4 · pi"]
end
phone -- "messages / voice" --> bridge
bridge -- "responses / live view" --> phone
bridge <--> machine
style phone fill:#e8f4fd,stroke:#0088cc,stroke-width:2px,color:#333
style bridge fill:#fff8e1,stroke:#f9a825,stroke-width:2px,color:#333
style machine fill:#f0faf0,stroke:#2ea44f,stroke-width:2px,color:#333
Each Telegram topic maps to one tmux window. With Herdr, it maps instead to one guarded agent session: agent.list is the sole identity source and CCGram persists only an opaque herdr-session-v1-… target, never a tab, pane, or terminal ID. Every Herdr agent topic is pane-qualified as <workspace> ▸ <tab> ▸ <pane>, so its label remains stable when siblings join or leave the tab. Every action reads a fresh agent.list record and fails closed for missing, malformed, sessionless, or legacy bindings. Duplicate canonical targets are quarantined while unrelated sessions remain operational. Legacy locator bindings require explicit rebind and are never inferred from names. A session can still change after that guard and before Herdr dispatches, so delivery is not atomic and may be indeterminate after this post-guard race.
What You Can Do
- Bind agents to topics — one agent per Telegram topic; create via directory browser
- Auto-detect providers — Supports Claude Code, Codex, Gemini, Pi, and Shell simultaneously
- Monitor live — Terminal screenshots on demand or auto-refresh every 5 seconds
- Send commands — Slash commands, voice messages (transcribed via Whisper), or raw shell input
- Run multiple agents in parallel — each topic independent; run different agents at once
- Recover gracefully — Resume, continue, or start fresh if a session crashes
- Send workspace files — Share files to Telegram via
/send(glob, path, or substring search) - Action toolbar — Provider-specific buttons for common actions (Screenshot, Mode, Esc, Enter, etc.)
Delivery and Sync Safety
CCGram losslessly combines only eligible consecutive transcript text deliveries for the same chat, topic, window, role, and source session. It preserves each item's formatting and keeps tool updates, media, status updates, and other boundaries separate. The status bubble shows queue progress; at a severe backlog (100 pending items or an oldest item aged 5 minutes), its inline Jump to live action requires confirmation and posts a skipped-range notice. The raw provider transcript is never deleted. Delivery is at-least-once, so a Telegram failure or restart before acknowledgement can repeat a transcript message rather than silently losing it.
/sync can clean up only locally recorded, eligible retired topics. It never discovers or enumerates arbitrary Telegram topics; an active or rebound topic is protected before any cleanup request. See the delivery, backlog, and Sync guide for boundaries, safety guarantees, and Telegram admin permissions.
Quick Start
Install:
uv tool install ccgram # recommended
# or: pipx install ccgram | brew install alexei-led/tap/ccgram
Telegram setup:
- Create a bot via @BotFather — full instructions
- Add bot to a Telegram group with Topics enabled; promote to Admin
- Create
~/.ccgram/.env:
TELEGRAM_BOT_TOKEN=your_bot_token_here
ALLOWED_USERS=your_telegram_user_id
CCGRAM_GROUP_ID=your_telegram_group_id
Get user ID from @userinfobot. Get group ID via @RawDataBot (prefix Peer ID with -100).
Run:
ccgram
Open your Telegram group, create a topic, send a message — directory browser appears. Pick a project directory, choose your agent (Claude, Codex, Gemini, Pi, or Shell), and you're connected.
Prerequisites: Python 3.14+, tmux or herdr, and one agent CLI. CCGram does not modify agent SDKs.
Herdr setup
CCGram supports Herdr socket protocols 14–20. Later and otherwise unknown protocol versions are attempted with a warning for forward compatibility; individual command failures still surface if the protocol is not usable. Telegram rate limiting uses a protected PTB adapter seam and is therefore tested against and constrained to python-telegram-bot>=22.6,<22.7. Install Herdr's integration before launching an agent that needs a native session identity:
herdr integration install pi
herdr integration install antigravity-cli
Restart an already-running agent after installation. Antigravity receives a native Herdr session identity after its first prompt creates a conversation.
Start new agents, or restart already-running agents, after installing the integration so they publish their agent_session identity. Then set CCGRAM_MULTIPLEXER=herdr and run ccgram hook --install as usual.
Platform Support
CCGram supports Linux, macOS, and WSL2. Native Windows is not supported.
On Windows, install and run CCGram inside WSL2. Install tmux or herdr and the agent CLI inside the WSL distribution.
Native Windows does not provide the Unix file locking, signal handling, and terminal multiplexer features that CCGram requires.
Documentation
- Guides — CLI reference, configuration, delivery/backlog safety,
/sync, voice transcription, multi-instance setup, session recovery, testing - Providers — Claude Code, Codex, Gemini, Pi, Shell; transcript delivery, session modes, LLM config, custom commands, git worktrees
- Architecture — delivery queue, transcript watermark, and provider/multiplexer design
Optional Features
Web Dashboard — Live terminal (xterm.js), transcript search, multi-pane grid in Telegram. Disabled by default. Enable here.
Development
git clone https://github.com/alexei-led/ccgram.git && cd ccgram
uv sync --extra dev
make check # lint, format, typecheck, test
make test-e2e # end-to-end tests (requires agent CLIs; see docs/guides.md#e2e-tests)
License
Comments (0)
Sign in to join the discussion.
No comments yet
Be the first to share your take.