Self-hosted · MIT · Python + Node

Claude AI
on your phone.

A bot that puts Claude on Telegram, WhatsApp, Slack and a browser chat — all from a single process. No webhook server. No public URL. No cloud relay.

npm npm install -g telechat
pip pip install telechatai

Then telechat init — an AI-guided setup that opens the right pages, collects your tokens and validates everything.
Runs free on your existing Claude Code subscription, or on an Anthropic API key.

Telechat · Claude Code CLI mode
Refactoring the payment module — 3 files staged.
Tool approval required
$ rm -rf dist/ && npm run build
✓ Approve ✕ Deny
Approve
Build passed in 12.4s. Session cost today: $0.34 / $5.00.
Desktop Bridge · approvals routed to your phone
One process, four surfaces
TelegramWhatsAppSlackWeb chat

What's inside

Not a wrapper. A control plane
for your Claude sessions.

Telechat runs as a background service on your own machine and keeps every message, memory and dollar of spend in a local SQLite database.

01

Claude Desktop Bridge

Hooks into your local Claude Code sessions. Get push notifications when Claude stalls or wants to run something destructive, inject replies from your phone, and gate every tool call behind an Approve / Deny tap.

02

Model registry

Every model ID and price lives in one module — Haiku, Sonnet and Opus tiers, each overridable with MODEL_SONNET and friends. A model bump is a config change, not a release.

03

Smart routing

Queries are classified by complexity and routed to the cheapest model that can answer them — Haiku for quick facts, Opus for architecture. Speed when it's cheap, depth when it counts.

04

Cost tracking & budgets

Per-user daily and monthly caps with warnings at 80%, computed from real token counts in API mode and reported spend in CLI mode. Check it any time with /budget.

05

Memory & sessions

Per-user memories with SQLite FTS5 full-text search, plus conversations you can create, switch, pin, archive and fork. Plan in one session, implement in another.

06

Bring your own providers

Claude drives the conversation; optional keys unlock the rest — OpenAI for speech and images, Replicate for music and video, Brave or Tavily for web search. Every one is opt-in.

Quickstart

Running in about sixty seconds.

1

Install

Global install from npm or PyPI — both ship the same telechat CLI.

2

Set up your platforms

telechat init drives the setup with Claude. No Claude CLI? Use telechat setup for a plain wizard.

3

Start the service

It survives terminal close and Ctrl+C. Manage it with start, stop, status and logs.

4

Or run it in Docker

API mode only — CLI mode needs the claude binary and its host auth, neither of which belongs in a container.

~/telechat — zsh 
# 1 — install (npm, or: pip install telechatai)
$ npm install -g telechat

# 2 — AI-guided setup: tokens, allowlists, .env
$ telechat init

# 3 — start as a background service
$ telechat
✓ Telegram bridge online
✓ Slack socket connected
✓ Claude mode: cli (subscription)

$ telechat status
telechat running · pid 48213 · uptime 2h14m

# .env — pick your surfaces
BOT_MODE=telegram,slack   # or: all
CLAUDE_MODE=cli           # or: api
COST_DAILY_BUDGET=5.0

# always-on server, API mode
$ docker compose up -d

CLI

Everything is a one-liner.

The bot runs as a background service. These are the controls.

telechatStart the bot as a background service
telechat stopStop the bot
telechat restartRestart the bot
telechat statusCheck whether it's running
telechat logsTail the bot log
telechat initAI-guided setup using the Claude CLI
telechat setupManual step-by-step setup wizard
telechat envShow environment variables, tokens masked
telechat env cleanRemove the .env file and clear all credentials
telechat updateUpdate to the latest version
telechat --debugStart with verbose logging

Get started

Put Claude in your pocket.

npm npm install -g telechat && telechat init
Read the docs npm ↗ PyPI ↗ Blog ↗