Get access

Deploy your own fleet of agents.

FleetKit lets your agents, like Claude or Codex, spin up a fleet in the cloud: 1, 10 or 100 of them. Each one runs on your own stack and your own subscriptions or keys, for free, and you can see what every agent is doing.


      

      

        
Let your agent set it up

        

      

    

  






  

  

    
Why we built  FleetKit

    
We run a lot of agents. Some write our pages, some test our apps, and some fix whatever the tests break. At some point one laptop wasn't enough. We wanted a whole fleet in the cloud, working while we sleep.

    
So we went shopping, and everything had a catch. Devin wants you on its stack. T3, Orca and OpenHands want you inside their desktop app. But we live in Claude Code and Codex, and we weren't moving out. We didn't need another app for us. We needed something for our agents.

    
The labs' own cloud agents looked like the easy way out, until we saw what a fleet costs at API prices. An arm and a leg. It also meant one model for everything, when we'd rather send research to Gemini, writing to Sonnet and the last review to Opus.

    
Meanwhile our Claude and ChatGPT plans sat half used every week. So we wired it all up ourselves: three or four API keys, three or four accounts, a pile of VMs to babysit. It worked. It was also a pain.

    
So we built the thing we wanted. With FleetKit your agents spin up a fleet on your own stack and your own subscriptions, at cost, and you can see what every one of them is doing. We'd love for you to try it!

  






  

  

    
A model for every job.

    
Each agent gets the model, harness and sandbox that fit its job, all in one fleet.

    

      
ResearchGeminiGemini CLI

      
WritingSonnetClaude Code

      
Final reviewOpusClaude Code

      
JudgingGPTCodex

    

  






  

  

    
Use what you already pay for.

    
Run on your own Claude and ChatGPT subscriptions, so the unused part of your 5-hour and weekly limits stops going to waste. FleetKit spreads the work across your accounts.

    

      
Claude Max · 1
80%

      
Claude Max · 2
70%

      
ChatGPT Pro
64%

      
you, by handyour fleet, in the gaps

    

    
Or bring an API key. Either way, we pass on the cost exactly, like OpenRouter.

    
$0on top. Ever.

  






  

  

    
See what every agent is doing.

    
The map shows each agent's live state. Your own Claude drives the same fleet over MCP: run, wait, message, stop.

    

      
agentstackstate

      
writeSonnet · Claude Coderunning

      
judgeGPT · Codexdone

      
fixOpus · Claude Codeasks you

    

  






  

  

    
Example: an SEO fleet.

    
Ten pages a day, each through three agents with three different models.

    

      
ResearchGemini · Gemini CLI
→
      
WriteSonnet · Claude Code
→
      
ReviewOpus · Claude Code

    

    
All on one GCP sandbox, billed at cost.

  






  

  

    
Example: a product that improves itself.

    
A loop that runs while you sleep, and stops to ask when it isn't sure.

    

      
Simulated userswalk your product
→
      
Judgea judge model scores it
→
      
FixOpus patches it
→
      
Re-checkrun it again

    

    
On Daytona, with each step in the harness that suits it.

  






  

  

    
Open source. Self-host it, or use ours.

    

      
1
Use oursAdd your keys or subscriptions and call fk.run(). You pay the providers' list price.

      
2
Self-host itRun the server on your own Google Cloud or AWS account. Your sandboxes, your bill, our code.

    

    
APACHE-2.0 Open source. The repo opens soon.

    

      
Get access

      
FleetKit is in private beta, free while it runs on our own cloud credits. Leave your email and we'll let you in.