Infrastructure-grade control for AI agents

Launch from your phone. Monitor from your browser. Rollback in one click.

The control plane your AI agents were missing. Snapshot every run, see what changed, share a live preview - and keep context across agents automatically.

Start free, upgrade anytime.

$ npm i -g cosmios
added 1 package in 4s
$ cosmios init
Opening COSMIOS in browser...
Machine approved. Connected.
Ready. Waiting for runs.
Auto snapshot before every runOne-click rollbackShare live previewsCross-agent memoryAutomated task queuesYour code never leaves your machine

The difference is night and day.

Without COSMIOS
  • SSH into server to check agent status
  • No idea what files were changed
  • Agent broke the build? Good luck
  • Context lost when switching agents
  • No way to preview what was built
  • Run tasks one by one, wait for each to finish
With COSMIOS
  • Control any machine from your browser
  • Every change tracked, every file diffed
  • One-click rollback to pre-run state
  • AI summaries carry context between agents
  • Live preview URL for every project
  • Queue missions, tasks chain automatically
cosmios.dev/dashboard
COSMIOS
Overview
Machines
Agents
Tasks
Backups
Logs
Usage
Workspaces
2
Machines Online
1
Agents Running
7
Tasks Today
0
Errors
Agents
Claude Code
Mac Studio
running
Codex
MacBook Pro
idle
Output
Agent
Adding OAuth provider...
Thinking...
$ Edit src/auth.ts
+ import OAuthProvider
You
Also add GitHub as a provider
Agent
On it. Adding GitHub OAuth...
Reply to agent...Send
Add OAuth login with Google
2Online
1Running
7Tasks
Claude Coderunning
$ Edit src/auth.ts
+ import OAuthProvider
You
Add GitHub too
Reply to agent...Send
HomeTasksMachinesIncidentsMore

Up and running in three commands.

No wrappers. No config files. No changes to your agent setup.

1

Install

npm i -g cosmios

One package. Works on macOS, Linux, Windows.

2

Connect

cosmios init

Approve in browser. Machine appears in your dashboard.

3

Control

cosmios.dev/dashboard

Dispatch tasks, monitor runs, rollback mistakes. From anywhere.

Everything your AI agent is missing.

No other tool gives you this. Not your terminal, not your IDE, not your agent.

REMOTE CONTROL

Launch agents from any browser. Even your phone.

Pick your agent, choose a workspace, type a task. Click Run. Watch it work in real time. Your machine can be anywhere - basement server, cloud VPS, office Mac. You just need a browser.

Claude Code
Sonnet 4
/home/deploy/my-saas
Add OAuth login with Google and GitHub providers
CROSS-AGENT MEMORY

Switch agents. Keep the context.

Every run is summarized. Switch from Claude Code to Codex - the new agent knows what the previous one did.

Run 1 - Claude Code
Added OAuth with Google provider, created auth.ts
Summary generated automatically
Run 2 - Codex
[Context injected: knows about auth.ts, OAuth setup]
Adding GitHub as second OAuth provider...
LIVE PREVIEW

Share what your agent built. One link.

Preview URL auto-generated per project. Send it to anyone - no deploy, no config.

my-saas.preview.cosmios.dev
My SaaS App
Welcome to the dashboard. Login | Register
Running on mac-studio via COSMIOS tunnel
SNAPSHOT & ROLLBACK

Every run gets a safety net.

Auto git snapshot before the agent touches anything. If it breaks your code, roll back in one click. Instant recovery.

[snapshot] commit a3f8c21
[snapshot] 14 files staged
...
[!] Agent broke build
> Rollback to a3f8c21? [Yes]
[rollback] HEAD reset. Build passing.
LIVE LOGS

See what your agent does. In real time.

Structured logs with diffs, tool calls, and edits. Not raw stdout dumped 3 hours later.

Agent
Looking at auth module. Adding OAuth provider...
$ Edit src/auth.ts (+48 -12)
+ import { OAuthProvider } from './oauth';
$ npm test
42 tests passed, 0 failed
RUN HISTORY

Every run documented. Automatically.

AI-generated summaries of every run. See what was done, which files changed, what decisions were made. Full audit trail.

What was done
- Added Google OAuth provider to auth.ts
- Created callback route in /api/auth
Files modified
- src/auth.ts, src/api/auth/callback.ts
Current state
- All tests passing. OAuth working.

Queue your tasks. Walk away.

Stack your instructions, hit Play, and let the agent handle the rest.

Pending
Set up Google OAuth provider
Claude Code
Pending
Add GitHub OAuth with callback
Claude Code
Pending
Create email/password auth flow
Codex
Pending
Write integration tests for auth
Claude Code
Mission completed. 4/4 cards done.
Automatically generated to-do list - things that need your manual action.
Configure GOOGLE_CLIENT_ID in .env
Set up OAuth callback URL in Google Console
Add GITHUB_CLIENT_SECRET to environment

Built for how you actually work.

Solo dev on a VPS

Your agent runs on a headless server. You monitor, dispatch, and rollback from your laptop or phone. No SSH needed.

On the go

Dispatch a task from your phone while commuting. Come back to a finished run with a full summary of what was done.

Multi-agent workflows

Start with Claude Code for architecture, switch to Codex for implementation. COSMIOS carries the context between them automatically.

Works with the agents you already use.

Claude Code, Codex and Gemini CLI supported today. Aider coming soon.

Claude Code
Anthropic
Codex
OpenAI
Ai
Aider
Open Source
Coming soon
Gemini CLI
Google

Simple, transparent pricing.

Start free. Scale when you need to.

MonthlyYearly-20%
Free
$0
  • 1 machine
  • 25,000 runs tracked/mo
  • 7 days history
  • Missions (5 cards/mission)
  • Remote dispatch
  • Rollback
  • URL Preview
  • Cross-agent memory
Start FreeNo credit card required.
Team
$99/mo
  • 10 machines
  • 1,000,000 runs tracked/mo
  • 90 days history
  • Missions (unlimited cards)
  • Remote dispatch
  • Rollback
  • URL Preview
  • Cross-agent memory
  • API & Webhooks
  • RBAC

You have questions. We have answers.

Which AI agents does COSMIOS support?

Claude Code, Codex and Gemini CLI are fully supported today. Aider is coming soon. COSMIOS works alongside your existing agent setup - no wrappers, no config files to write.

Does COSMIOS have access to my source code?

No. The server only receives metadata, logs, and commit hashes. Source code never leaves your machine. API tokens are SHA-256 hashed at rest. All communication over TLS.

How does snapshot and rollback work?

Before every run, the COSMIOS runner executes git add -A && git commit automatically. If the agent breaks something, click Rollback in the dashboard and the runner resets to the snapshot commit. Instant recovery.

Do I need a VPS?

No. COSMIOS works on any machine - your laptop, a Mac Mini, or a headless VPS. Install with npm i -g cosmios, run cosmios init, and control your agents from any browser.

Is COSMIOS free?

Yes. The free plan includes 1 machine, automatic snapshots, one-click rollback, and preview URLs. Start free, upgrade when you need more machines or history.

How does cross-agent memory work?

After each run, COSMIOS generates a structured summary of what was done. When you start a new run on the same project - even with a different agent - the summary is automatically injected into the prompt. Codex picks up where Claude Code left off, and vice versa. No manual context passing needed.

Can I change model or agent options remotely?

Yes. The dashboard lets you pick the model, set max turns, enable skills, and filter tools. No SSH, no config files. Everything from the browser.

What happens if my machine goes offline?

The agent continues running locally. Events are journaled and automatically synced when the connection is restored. The dashboard shows the machine as offline and you will see the full log when it reconnects.

Stop babysitting your AI agent.

Set up in 2 minutes. Free plan available.

npm i -g cosmios
cosmios init
Start building in 2 minutes

Start free, upgrade anytime.