Free to start · No credit card required

The operating system for AI agents

Monitor every run. Host agents without infrastructure. Get alerted when things break. Ship AI agents with the confidence of a production system.

Works with any language or runtime — Python, Node.js, Go, or plain curl

email-classifier · run_7f3alive
00agent.run.started
01llm.request
02tool.called
03tool.returned
04llm.response
05agent.run.completed
06
6
events
1
tool call
1.2s
duration

How it works

Two ways to get started

Host a new agent entirely in AgentOS, or drop the SDK into your existing code. Both give you full observability.

1

No agent yet

AgentOS runs your agent

Define your system prompt, tools, and model in the dashboard. AgentOS handles the LLM calls, tool execution, and the full run loop. You invoke it with a single HTTP call and get back the output.

// Your entire agent is one HTTP call
fetch(`…/agents/${AGENT_ID}/invoke`, {
  method: "POST",
  headers: { "Authorization": `Bearer ${API_KEY}` },
  body: JSON.stringify({ input }),
});
2

Already have an agent

AgentOS observes your agent

Your agent code stays exactly where it is — your server, Lambda, Colab, anywhere. Add 3 SDK lines to stream every run, event, and output to AgentOS. We record it, you watch it.

// Your existing agent — 3 lines added
const run = await aos.startRun({ input });
/* ... your existing agent logic ... */
await run.complete({ output });  // streamed to AgentOS

Your code → AgentOS → full observability

Your Codeany languageSDK / APIstartRun · emitAgentOSruns + alerts

Features

Everything your agents need to ship safely

Real-time observability

Every event, tool call, and LLM response captured with millisecond precision. Live run timeline. Never debug blind again.

Hosted agents

Define system prompts, tools, and models. Run agents directly from AgentOS — no infra, no cold starts, no ops.

Human-in-the-loop approvals

Gate any tool call behind a human approval. The agent pauses, notifies your team, and waits — no run gets through without sign-off.

Smart alerts

Set conditions on error rate, duration, silence, or custom events. Get notified via in-app, email, or webhook before your users notice.

AI reports

Daily and weekly AI-generated summaries of agent performance. Understand patterns across thousands of runs at a glance.

MCP server

Connect Claude Desktop, Cursor, or any MCP client to your workspace. Manage agents, runs, and reports from inside your IDE.

import

TypeScript & Python SDKs

Instrument any agent in two lines. startRun(), emit(), endRun() — full observability from your existing code, any runtime.

Ingest API & webhooks

POST events from any language or runtime via the ingest API. Receive webhook notifications when runs complete, fail, or need attention.

Agent marketplace

Install pre-built agents in one click. Share your own templates with the community. Skip the blank page.

Install in seconds. Works with TypeScript, Python, or any HTTP client.

$ npm install @agentos-sdk/core$ pip install pyagentos-sdkRead the docs →

Pricing

Start free, scale when ready

No credit card required to get started.

Free
$0forever

For solo developers and side projects.

  • 2 agents
  • 1,000 runs / month
  • In-app alerts
  • Community support
Get started
Most popular
Team
$20/ seat / month

For companies running agents together.

  • Unlimited agents
  • Teams & per-folder access control
  • Invite your whole company
  • By-employee activity view
  • All integrations + human approvals
  • Email support

or $200/seat/year billed annually — 2 months free

Start with Team
Enterprise
Custom

For companies running agents at scale.

  • Everything in Team
  • SSO / SAML & SCIM
  • Audit logs & data retention
  • IP allowlisting
  • SLA + dedicated support
Contact us

FAQ

Common questions

Do I need to rewrite my agent to use AgentOS?+

No. If you have an existing agent, add our SDK (npm or pip) and wrap your code with startRun/complete. Takes about 10 minutes. Your agent keeps running wherever it is — AgentOS just observes it.

What's the difference between hosted and non-hosted agents?+

Non-hosted: your code runs the agent, AgentOS observes it. Hosted: you define the prompt and tools inside AgentOS and we run it for you. Both give you the same dashboard, alerts, and reports.

Is my run data private?+

Yes. Run data is scoped to your workspace and only accessible by workspace members. We do not use your run data to train models.

Can I use AgentOS with Claude, GPT-4, or any other model?+

For hosted agents, AgentOS supports Anthropic and OpenAI models. For non-hosted agents, your code picks the model — AgentOS doesn't care, it just observes the run.

What happens when I hit the free plan run limit?+

Runs that exceed the monthly limit are dropped. You'll get a warning notification before you hit the cap. Upgrading to Team removes the limit.

Start monitoring your agents
in under 5 minutes.

Free forever for solo developers. No credit card, no lock-in.