01
Watch it think, live.
Runs stream in event by event: the prompt, every tool call, every model reply, with tokens and cost counting up as they happen. No refresh, no tailing logs.
Observability and control for AI agents
Agents do the work.
You see every move.
Everyone is building agents. Almost no one can run them safely. AgentOS puts every run on the record, and a hard pause for your approval before anything sensitive happens.
agentos · overview● live
142
runs · 7 days
92%
success · 7 days
11
failures · 7 days
1.7M
tokens · 7 days
completed
awaiting approval
running
completed
failed · alert sent
100%
100%
93%
88%
Works with the models and tools your agents use
- OpenAI
- Anthropic
- Gmail
- Slack
- GitHub
- Google Sheets
- Notion
- HubSpot
- Linear
- 6 agents run our own company
- Zero markup on your model tokens
- Approval gate on any tool
- Self-hosted option
A day in our workspace, replayedReplay
Observability
What did the agent do?
The answer is one click away.
AI agent observability, done properly: every run streams live, lands on the record in full, and stays watched after you look away.
every event
on a millisecond timeline
every token
in and out, per run
every decision
who approved what, and when
02
Replay any run to the millisecond.
Inputs, outputs, tool calls, tokens, cost and every approval decision, kept for every run. When someone asks what the agent did, the answer is a link, not a meeting.
03
Hear about it before anyone asks.
Failures, silence and slow runs raise an alert. Intelligence agents audit the fleet for regressions and wasted tokens, and file the finding with the fix.
Human in the loop
Nothing sensitive happens without a human.
Mark any tool as gated: sending email, posting to social, writing to a spreadsheet. The run pauses, the action lands in your approvals inbox exactly as it will be sent, and a batch becomes one decision.
approvals · outbound-outreachpaused
Outbound Outreach wants to send 3 emails.
- To
- andreia@northwind.example
- From
- bernabranco@agentos-ai.dev
- Subject
- Your support queue, answered overnight
Hi Andreia, Saw Northwind is hiring a second support lead. We run an agent that drafts replies to routine tickets overnight, and a person approves anything sensitive before it goes out. Worth a 20-minute look next week? Bernardo
Details
{
"recipient_email": "andreia@northwind.example",
"from_email": "bernabranco@agentos-ai.dev",
"subject": "Your support queue, answered overnight",
"body": "Hi Andreia,\n\nSaw Northwind is hiring a second support lead. We run an agent that drafts replies to routine tickets overnight, and a person approves anything sensitive before it goes out.\n\nWorth a 20-minute look next week?\n\nBernardo"
}For developers
Already built agents? Keep them where they are.
A few SDK lines or an OpenTelemetry export puts any agent, in any language, on the same record as hosted agents: runs, traces, token costs, alerts and approvals.
$ npm install @agentos-sdk/core$ pip install pyagentos-sdkagent.tsTypeScript
import { AgentOS } from "@agentos-sdk/core";
const aos = new AgentOS();
await aos.withRun(async (run) => {
const step = run.startSpan("classify ticket");
run.emitToolCall("lookup_order", { orderId: 991 });
step.end();
});Proof, not promises
Some of the agents that run our company
These aren't demos. Our inbox, support queue, outreach, social feed and on-call briefings: agents do the work, humans approve what matters.
Email Triage
runs nightlyReads every inbound email, classifies it (support, sales, billing, spam) and files it with the right label. Genuinely ambiguous mail stays in the inbox for a human.
Support Reply
runs nightlyDrafts grounded replies to support email and sends the routine ones. Complex or sensitive threads become drafts for human review instead of auto-sending.
Outbound Outreach
on demandSends personalised outreach from the prospect sheet, tracks what was sent, and surfaces replies for a human to handle.
Social Content
runs MondaysDrafts a week of LinkedIn and X posts every Monday and emails them for review. Publishing is approval-gated: nothing goes live without a human clicking approve.
Competitor Intel
runs MondaysFetches competitor blogs and changelogs, summarises what each one shipped last week, and emails a structured intel brief.
Incident Monitor
runs morningsSummarises overnight alerts and incidents into a single what-happened-what's-open brief for whoever is on call.
Features
Real work needs real guardrails
Smart alerts
Conditions on error rate, duration, silence, or custom events. In-app, email, or webhook. You hear about problems before your customers do.
AI reports & intelligence
Built-in intelligence agents watch your agents for anomalies, regressions, prompt weaknesses, and token waste. Daily AI summaries keep you ahead of it.
Hosted agents
Define the job, tools, and model in the dashboard. AgentOS runs the whole loop: no infra, no cold starts, no ops. Invoke with one HTTP call or a schedule.
Agent marketplace
Install proven agents in one click, including the ones running our own company. Skip the blank page; start from something that already works.
Bring the agents you already have
Instrument code agents in three lines with the TypeScript or Python SDK, POST events from any language, or point your OpenTelemetry GenAI traces at our OTLP endpoint.
MCP server
Connect Claude, Cursor, or any MCP client to your workspace. Manage agents, approve runs, and read reports without leaving your editor.
Pricing
Pay for agents. People are free.
Cheaper than the work it replaces. Two weeks of pilot cost nothing.
Solo
€39/ month
You build and run the agents yourself.
- 3 agents included, then €9 per agent
- Unlimited members: we don't charge for people
- Human approval gates on any tool
- All integrations: Gmail, Sheets, GitHub & more
- Email support
or €390/year billed annually (2 months free)
Operations
€199/ month
We build and run up to 3 agents for you.
- Up to 3 agents, implemented by us
- Monitoring, fixes and a monthly report
- No per-agent charge
- Everything in Solo
Initial implementation quoted separately
EvolutionMost popular
€499/ month
Up to 8 agents, plus our work every month.
- Up to 8 agents, implemented by us
- 4 hours of our work each month for changes and new flows
- Priority support
- Everything in Operations
Initial implementation quoted separately
Enterprise
Custom
For companies running agents at scale.
- Everything in Evolution
- Self-hosted / on-prem option
- SSO / SAML & SCIM
- Audit logs & data retention
- SLA + dedicated support
AI usage is bring-your-own-key: agents run on your OpenAI, Anthropic, or any compatible endpoint, billed by the provider at their normal price. AgentOS takes no markup on your tokens, ever.
FAQ
Common questions
Do I need to be a developer to use AgentOS?
No. Install an agent from the marketplace, connect your tools with OAuth, and it starts working. Writing your own agent is describing the job in plain language: the prompt, the tools it may use, and which of those need your approval. Developers get SDKs and APIs on top, but they're optional.
What does AI agent observability actually mean here?
That nothing an agent does is invisible. Every run is streamed live and stored in full: each tool call, each model response, tokens, cost, and any human approval decision, with millisecond timestamps. You can watch a run as it happens, replay any past run step by step, and set alerts on failures, silence, or slow runs. It works for hosted agents automatically, and for agents you already run elsewhere via the SDK or OpenTelemetry.
Can an agent do something risky without my sign-off?
Not if you gate it. Any tool can be marked as requiring approval, whether it sends email, posts to social, or writes to a spreadsheet. A gated call pauses the run, lands in your approvals inbox, and nothing happens until a human approves or rejects it. We run our own social publishing this way.
What actually runs the agent? Do I need servers?
AgentOS hosts and runs your agents: the LLM calls, the tool executions, the schedules, the retries. There's nothing for you to deploy or keep alive. If you'd rather run agents in your own code, the SDK reports them into the same dashboard.
Do I need to rewrite agents I already have?
No. Add the SDK (npm or pip) and wrap your code with startRun/complete. It takes about 10 minutes. Or export OpenTelemetry traces to our OTLP endpoint with zero AgentOS-specific code. Your agent keeps running wherever it is.
Is my data private?
Yes. Run data is scoped to your workspace and only accessible by workspace members. Credentials for connected tools are encrypted at rest, and secrets are readable only by workspace admins. We do not use your data to train models. An on-prem / self-hosted option exists for Enterprise.
Which models can agents use?
Hosted agents run on Anthropic and OpenAI models out of the box, plus any OpenAI-compatible endpoint (including self-hosted models on Enterprise). Agents you run through the SDK can use anything. AgentOS supervises the run either way.
We already pay for ChatGPT. Isn't this the same thing?
Different job. ChatGPT is people chatting with AI, and it's good at that. AgentOS is software doing jobs on its own: scheduled, supervised, auditable, with a human gate on anything sensitive. Many teams keep both: seats for the humans, AgentOS for the workforce of agents. What we'd argue against is paying per-person prices to run software.
What does it cost to start?
A 30-minute conversation, then a 2-week pilot at no cost. You bring one annoying, repetitive workflow (reporting, triage, chasing, monitoring); we design and implement the agent for it, with an approval gate on the scary part. At the end you get a report with real numbers: runs completed, time saved, cost per run. Plans start at €39/month if you want to build it yourself, or €199/month with us running it for you.
Give one job to an agent this week.
Install one of the agents that runs our company, gate the scary part behind your approval, and watch the first supervised run.