Built from one paragraph
Describe the job in plain text from Claude, Codex or Cursor. The agent gets models, tools, memory and files through one MCP.
Platform / Serverless agents
Serverless agents.
Fast to build. Cheap to run.
Describe an agent in plain text from Claude, Codex or Cursor. It runs in the FlyMy.AI cloud with every model and tool behind one MCP, and comes back as a versioned endpoint you can call, schedule or embed.
1500+ integrations500+ modelsup to 12x lower modeled run cost
Modeled agent-run cost: up to 12x lower
One fixed workload, priced at each platform's published rates. A cost model, not a benchmark.
12xmodeled gap
What a serverless agent gets here
Models, tools, compute, versions and schedules in one account. No servers, no seats.
Every tool behind one MCP
1500+ integrations, 500+ models, a real browser and an isolated sandbox, with the accounts you connect once.
Freeze a run into a version
Accept one run and freeze it. Frozen runs make repeat jobs cheap and fast; the live version rolls forward or back.
Runs on a schedule
A cron and a time zone, in the cloud with your laptop closed. Three failed runs in a row pause the schedule and notify you.
An endpoint for your product
Call a frozen agent by REST, SDK or MCP, or publish it as a deployment: your customers connect their own accounts and the runs bill to you.
A lead with subagents
A lead agent starts up to 6 subagents, up to 3 at a time, within a dollar cap you set.
Budgets and approvals
Give a run or a project a budget. Put an approval before a send, a payment or a delete.
State and history in the cloud
Every run keeps its context, files, tool calls and cost. Continue it from your coding agent.
From a paragraph to an endpoint
Four calls from your coding agent, or the same steps in the app.
- 01
Connect
Paste the guide link https://flymy.ai/skill.md into Claude, Codex or Cursor, or add the MCP by hand.
- 02
Describe and run
create_agent with the job in plain text, then run_agent once and read the result and its cost.
- 03
Freeze
freeze_agent on the run you accept, then test that version with run_frozen.
- 04
Schedule or call
schedule_agent with a cron and a time zone, or call the frozen version by API from your product.
agent = create_agent({"name":"Health check","mcp_access_mode":"legacy",
"user_prompt":"Return a JSON object with status set to ok."})
run = run_agent({"agent_id":agent.uuid,"variables":{},
"message":"Run the health check now.",
"operation_key":"health-run-<uuid>"})
page = get_run({"execution_id":run.id}) # repeat with since=page.next_since
frozen = freeze_agent({"execution_id":run.id})
# Poll get_compilation until compiled; stop on failure.
test = run_frozen({"compilation_id":frozen.id,"operation_key":"health-frozen-<uuid>"})Pay for what a run uses
Model tokens, tool seconds, browser and sandbox time. No seats, no minimum, no subscription required.
Questions
What is a serverless agent?
An agent that runs in the cloud when it is called, by API or on a schedule, with no server for you to manage. You describe it, test one run and freeze it; FlyMy.AI runs that version every time it is called.
Where does the 12x come from?
From a cost model of one fixed workload: 8 model steps, 4 web searches, 2 minutes of browser and 90 seconds of sandbox, priced at each platform's published rates on 5 October 2026. FlyMy.AI with an open model came to $0.029 a run and OpenAI Managed Agents to $0.351. The platforms use different models and tool providers and task quality was not compared, so read it as an estimate, not a benchmark.
Can I use Claude or GPT models?
Yes. You choose the model for each agent, open or proprietary; the second bar above is Claude Sonnet 5 running on FlyMy.AI.
Do I need my own API keys?
Not for models and the shared connections: they bill to your FlyMy.AI balance. A service that acts as you, such as your own Google or Telegram account, is connected once on its own page. Keys never go into the chat.
Go to the source
$ start
Describe your first agent.
Free to sign up. No seats, no minimum - you pay only when your agents run.
Checked 2026-10-28. Features follow the agent guide and the docs; the cost chart is a model from published rates on 5 October 2026.