pochi 0.1.3

Multi-model AI agent Telegram bot for multi-folder workspaces.

pochi

Multi-model AI agent Telegram bot for multi-folder workspaces.

Run AI coding agents (Claude Code, Codex, OpenCode, Pi, etc.) from Telegram with workspace-based organization. Each folder gets its own Telegram topic, enabling parallel work across multiple projects from a single bot.

features

• multi-engine: supports multiple AI agent backends (Claude, Codex, OpenCode, Pi) via auto-discovery
• workspace mode: organize multiple folders under one Telegram group with topic-based isolation
• stateless resume: continue a thread in chat or pick up in the terminal with claude --resume <token>
• progress updates: real-time streaming of commands, tools, file changes, and elapsed time
• ralph wiggum loops: iterative Claude prompting where it reviews its own work until complete
• folder management: clone repos, create folders (with or without git), or add existing directories from Telegram
• robust rendering: markdown output with quality-of-life tweaks for Telegram

requirements

• uv for installation (curl -LsSf https://astral.sh/uv/install.sh | sh)
• python 3.14+ (uv can install it: uv python install 3.14)
• claude CLI on PATH (npm install -g @anthropic-ai/claude-code)
• a Telegram bot token (from @BotFather)
• a Telegram group with forum/topics enabled

install

uv python install 3.14
uv tool install -U pochi
# or try it with
uvx pochi@latest

quick start

1. create a Telegram bot via @BotFather and save the token

2. create a Telegram group with topics enabled:

   • create a new group
   • go to group settings → Topics → enable
   • add your bot to the group as admin

3. get the group ID: the easiest way is to add @RawDataBot to the group temporarily, it will show the chat ID (remove it after)

4. initialize a workspace:

   mkdir my-workspace && cd my-workspace
   pochi init
   # or non-interactive:
   pochi init --bot-token "123:ABC" --group-id -100123456789
   

5. run pochi:

   pochi
   

6. add folders from the General topic in Telegram:

   /clone backend [e-mail]:user/backend.git
   /create frontend
   /create knowledge-vault --no-git
   /add existing ~/dev/my-project
   

each folder gets its own topic. switch to a folder's topic to work on it.

workspace config

config lives at .pochi/workspace.toml:

[workspace]
name = "my-workspace"
telegram_group_id = -100123456789
bot_token = "123456789:ABCdefGHIjklMNOpqrsTUVwxyz"
default_engine = "claude"  # Optional, defaults to "claude"

[folders.backend]
path = "backend"
topic_id = 123
origin = "[e-mail]:user/backend.git"

[folders.frontend]
path = "frontend"
topic_id = 456

[folders.knowledge-vault]
path = "vault"
topic_id = 789
description = "Team knowledge base"

[workers.ralph]
enabled = false
default_max_iterations = 3

# Per-engine configuration sections
[claude]
model = "opus"
allowed_tools = ["Bash", "Read", "Edit", "Write"]

Folders can be git repos or plain directories. The type is auto-detected based on .git presence.

commands

CLI

pochi init [folder]     # initialize workspace (prompts for bot token + group ID)
pochi                   # run the bot
pochi info              # show workspace info
pochi --version         # show version
pochi --debug           # enable debug logging
pochi --no-final-notify # edit progress message instead of sending new final message

Telegram (General topic)

command	description
/clone <name> <url> [path]	clone a git repo and create a topic
/create <name> [--no-git]	create a new folder (with git init by default)
/add <name> <path>	add an existing folder to the workspace
/list	list all folders in the workspace
/remove <name>	remove folder from workspace (keeps files)
/status	show workspace status
/engine	show current default engine and available engines
/engine <name>	set default engine for new conversations
/help	show available commands

Telegram (folder topics)

command	description
/ralph <prompt> [--max-iterations N]	start iterative Claude loop
/cancel	cancel the current run or ralph loop

or just send a message to chat with your AI agent!

ralph wiggum loops

ralph loops run Claude iteratively, asking it to review its own work after each iteration. Claude signals completion by responding with RALPH_COMPLETE: <summary>.

/ralph-loop implement user authentication with tests --max-iterations 5

enable always-on ralph mode in config:

[workers.ralph]
enabled = true
default_max_iterations = 3

resume tokens

every response includes a resume line:

`claude --resume abc123`

• reply to a message with a resume line to continue that thread
• copy the command to resume interactively in your terminal
• resume tokens are scoped to topics: topic:123:abc123

notes

• the bot only responds to messages in the configured group
• run only one pochi instance per bot token
• run claude once interactively in each folder to trust the directory

acknowledgments

Built on top of takopi.

license

MIT