v0.11.0 · Multi-agent · Local-first

One command center for every AI coding agent.

Run Claude Code, Codex, Gemini, OpenCode, and IDE-based workflows from one local dashboard. Monitor sessions, approve tools, resume context, and steer work from desktop, browser, or Telegram.

Download Free See Telegram Control

Claude Code Codex Gemini OpenCode

6+ Concurrent Sessions

4 Agent Adapters

22 Telegram Commands

100+ Library Auto-Detect

Claude Refactor auth flow VS Code · working

Live

Codex Review Rust WS patch Terminal · waiting approval

Gate

Gemini Generate test matrix JetBrains · queued

Idle

/sessions

4 agents running. Codex needs permission.

Approve cargo test?

Approve Deny

Works across your coding stack

Claude Code Codex CLI Gemini CLI OpenCode VS Code Cursor JetBrains Terminal

One control plane

Everything your agents need,
nothing you have to babysit.

Run in parallel

Multi-agent session grid

Claude, Codex, Gemini, and OpenCode running side by side — each with its own terminal, adapter badge, live status, and quick controls.

Leave your desk

Telegram remote

A real remote, not a notification toy. Start, resume, approve tools, and hand a live thread between web and phone with Stream Bridge.

Permission gate

Approve or deny every tool call from web or Telegram. Color-coded danger flags keep you in control.

Magic Ring

A glowing orb for fleet health at a glance. Link sessions, share context, spark a debate.

Smart sessions & auto-compact

Auto-compact in 3 modes, cost budget caps, enhanced resume with search, and a context meter on every session.

Debate mode

Pit agents against each other in structured rounds — a verdict before code ships.

Docker or native

One docker compose up, or a Tauri desktop app with system tray + auto-updater.

Intelligence Layer

Your agents just got
a lot smarter.

Companion doesn't just manage sessions — it auto-injects code structure into every message. Agents skip the searching and start solving.

4+L0-L3 Context layers auto-loaded

~5 Search calls saved per message

74K → 20K Session context shrunk by Wiki

CodeGraph

Semantic Code Intelligence

4 Auto-Injection Points Project map at session start (~1,500 tokens). Related symbols per message (~800 tokens). Plan review catches missed deps. Break check warns before regressions.
5 Languages, 15 Symbol Types TypeScript, Python, Rust, Go, Java. Functions, classes, interfaces, hooks, endpoints — all with trust-weighted edges and signatures.
Prevents Regressions, Not Just Maps Code When your agent plans file changes, CodeGraph flags unmentioned dependencies. When files are modified, it warns which imports may break. Saves ~5 search calls per message.

Wiki KB

Living Docs for Agents

4-Layer Lazy Loading (L0 → L3) L0 index always injected (~500 tokens). L1 rules when hot. L2 patterns on demand. L3 examples only when the agent asks. Full 74K knowledge base, 20K in-context.
Cross-Adapter — Claude, Codex, Gemini, OpenCode One wiki feeds every CLI. Tool name rotates per adapter (companion_wiki_note for Claude, MCP for the rest) so agents record discoveries as they work.
Raw → Polished Pipeline Agents dump raw notes; a background LLM pass deduplicates, tags, and promotes them to structured articles. Your team's learnings compound.

Pricing

Simple, honest pricing.
Pay yearly, cancel anytime.

No hidden fees. Free tier is forever. Paid plans unlock more sessions and premium intelligence features.

You bring your own AI provider accounts and local CLIs. Companion is the management layer — we do not resell model access or touch provider billing.

Free

$0 forever

No account required

2 concurrent sessions
Web dashboard + Magic Ring
1 Telegram bot
Claude, Codex, Gemini, OpenCode adapters
Templates + free model debate
Desktop app (macOS/Win/Linux)
MCP auto-detect
14-day Pro trial included

Download Free

Recommended

Pro

$5 /month

or $39/year (save 35%)

Unlimited sessions
Multi-bot Telegram
Multi-platform debate (Codex, Gemini, OpenCode)
CodeGraph (blast radius, flow trace)
Custom domain + Let's Encrypt SSL
Custom personas + RTK Pro
30-day money-back guarantee

Secure checkout via Polar.sh 30-day money-back guarantee macOS · Windows · Linux

Get Started

Deploy in seconds.

Companion runs locally on your machine via Docker or desktop app. It manages supported coding CLIs and IDE workflows using your own provider accounts. Your code never leaves your computer.

bash

# 1. One-line install (downloads docker-compose.yml + .env)
curl -fsSL https://companion.theio.vn/install | sh
cd companion

# 2. Launch — free tier starts instantly
docker compose up -d

# ✓ Open http://localhost:3580
# ✓ 2 sessions free, forever
# ✓ Bring your own Claude/Codex/Gemini/OpenCode accounts

Requires Docker Desktop + at least one supported coding agent account or local CLI. Image pulled from ghcr.io.

powershell

# 1. One-line install (PowerShell)
irm https://companion.theio.vn/install.ps1 | iex
cd companion

# 2. Launch — free tier starts instantly
docker compose up -d

# ✓ Open http://localhost:3580
# ✓ 2 sessions free, forever
# ✓ Bring your own Claude/Codex/Gemini/OpenCode accounts

Requires Docker Desktop for Windows + at least one supported coding agent account or local CLI.

Native desktop app — no Docker required. Built-in server and auto-updater.

macOS (.dmg) Windows (.exe) Linux (.deb)

v0.11.0 · Bring your own supported coding agent accounts or local CLIs.

1 Open Settings in Companion web UI

2 Paste your license key in the License section

3 Click Activate — Pro unlocks instantly, no restart needed

License key emailed after payment. You can also set COMPANION_LICENSE in your .env file.

FAQ

Frequently asked questions.

Do I need my own AI provider accounts?

Yes. Companion uses your existing provider accounts and local CLIs. For Claude Code sessions, bring your own Claude subscription or API key. For Codex, Gemini, and OpenCode, configure the matching CLI/profile yourself. Companion is the management layer on top.

What does Companion actually do?

Companion manages supported AI coding agents on your machine: Claude Code, Codex, Gemini, OpenCode, and IDE-based workflows. It gives you a web dashboard, Telegram remote control, CodeGraph intelligence, real-time activity monitoring, and shared context via the Magic Ring — all from localhost.

What's the difference between Free and Pro?

Free forever gives you 2 concurrent sessions, the full dashboard, Magic Ring, 1 Telegram bot, and every agent adapter — no account, no card, plus a 14-day Pro trial. Pro ($5/mo or $39/yr, save 35%) unlocks unlimited sessions, multi-bot Telegram, cross-agent debate, CodeGraph blast-radius + flow trace, custom domain with SSL, custom personas, and RTK Pro compression. Cancel anytime.

What is CodeGraph?

CodeGraph scans your codebase across 5 languages (TypeScript, Python, Rust, Go, Java), builds a trust-weighted dependency graph, and auto-injects at 4 points: project map at session start, related symbols per message, plan review (catches missed dependencies), and break check (warns before regressions). Saves ~5 search calls per message.

Why does it run locally?

Coding agents and IDE workflows need direct filesystem access, so Companion runs on your machine via Docker or the desktop app at localhost:3580. The only external call is license verification via pay.theio.vn. Your code and credentials never leave your computer.

What are the system requirements?

Docker Desktop (recommended) or the native desktop app. Works on macOS, Linux, and Windows (WSL). Minimum 2GB RAM. You'll also need at least one supported coding agent account or local CLI configured.

Is there a money-back guarantee?

Yes. All paid licenses come with a 30-day money-back guarantee. If Companion isn't right for you, email us and we'll refund your purchase — no questions asked.

Ready to run your agent fleet from one place?

Start local. Add agents when you need them. Keep control from web, desktop, and Telegram.

Download Free Docker or Desktop · No account needed · 2 sessions forever