Your agents already do the work. Your company wasn't built for them.
AgentArea gives every agent its own access, budget and approvals, and carries long work through to the end. In our cloud or inside your perimeter.
Company infrastructure was built for people
Access, card limits, approvals and change logs assume an employee who works office hours and answers for what they do. An agent can run a hundred times overnight, touch a dozen systems and spend money at every step. It needs the same limits, in a different shape.
For a person
For an agent in AgentArea
- A job title and dutiesA role: its own instruction, model, tools and skills
- Access through SSOPermissions in a graph, checked before every tool call; access closes if OpenFGA is unavailable
- Passwords the employee keepsKeys in a secrets store, resolved on the server; the model never sees them
- A card with a limitA budget per run and per month: warning at 80%, calls blocked at 100% when the check succeeds. A failed budget check lets the call through
- Sign-off from a managerA risky step waits in your inbox for as long as it takes, and survives a restart
- A report on the work doneEvery task's history: steps, tool calls and what it cost
- A change logA record of every configuration change: who made it and exactly what changed
One answer for everyone who signs off
Putting agents to work is rarely one person's call. Here is what each of them asks first.
Founders and operators
How do I stop being the router between my agents?
- Agents hand work to each other and wait for the result
- You step in only where you set an approval
- Work picks up after a crash instead of dying overnight
Architects and platform teams
Why not assemble this ourselves?
- Durable execution, MCP gateway and policies in one place
- API, CLI and MCP for configuring agents and connecting harnesses
- Open source under Apache-2.0, Docker Compose or Helm
Security
What can an agent do, and what did it do?
- Permissions checked before every tool call
- Secrets resolved on the server, never shown to the model
- Task tool calls and configuration changes on record
Finance
What will it cost, and can the bill surprise us?
- A cap per run and per month for every agent
- Warning at 80%; calls blocked at 100% if the check succeeds. Failed budget checks let calls through
- The cost of every task, by agent
Start from a ready team
Typical teams, assembled from agents, with the limits already set: what each agent may do, and what waits for a person.
Takes requests, prices projects from their documents, collects supplier prices, sends letters only with the chief engineer's approval.
- Request intakeTelegram
Collects the brief into the project folder
- Estimator
Quantities and prices by the bureau's rules
- Procurement
Current supplier prices in one table
Has access to: Web
- Dispatch
Sends estimates and price requests
Has access to: Gmail
- Weekly digestFridays
Requests, estimates, what is blocked
Boundaries
- Every letter waits for the chief engineer
- Only Dispatch has access to mail
- Configured budget: $5 per estimate, $200 a month for the bureau
- No agent can delete files
Installs into your workspace; you can change any agent or limit afterwards. These are configured budgets, not price estimates. Failed budget checks let calls through.
How to start
Three ways in. Start with whatever you already have open.
1
In the browser
Create an agent, connect its tools and start the first task. No code.
Get early access2
Over MCP, CLI or REST
Most of what the interface does is available over the API: agents, runs, policies, secrets.
Copy commands
# MCP: Claude Code, Cursor or any MCP client$ claude mcp add --transport http agentarea https://api.agentarea.ai/mcp# CLI$ npm install -g @agentarea/cli# REST$ curl https://api.agentarea.ai/v1/workspaces/$WORKSPACE/agents \-H "Authorization: Bearer $AGENTAREA_API_KEY"
3
With your agent
Paste one prompt into Claude Code, Codex or Cursor. It connects AgentArea and sets up the agent for you.
Copy prompt
Connect the AgentArea MCP server at https://api.agentarea.ai/mcp and sign in when asked. Read https://docs.agentarea.ai/llms.txt, then create an agent in my workspace for: <describe the task>.
Where it runs
The same open-source product, in three places.
Our cloud
The managed cloud opens market by market. Leave your address and we'll write when it opens where you are.
Your perimeter
A separate Helm chart moves MCP servers, and optionally sandboxes, into your own cluster, under gVisor and NetworkPolicy.
Self-host
Open source under Apache-2.0. Docker Compose or Helm, your own Postgres, Redis, S3 and model endpoint.
Questions
What people ask before putting agents to work.
Give your agents the limits
you'd give a person
Access, a budget, sign-off on the risky steps and a record of what was done. Start in the cloud, or talk to us about your own infrastructure.