Agents that know your business
Give each agent a job, a model and a reading list. PDFs, sites and notes become searchable knowledge it cites back.
AiOrch is the control room for your company’s AI. Build agents and workflows without code, plug them into the tools you already run, and keep a person in the loop for anything that writes, pays or ships.
What the person writes in chat, plus who they are.
billing.find_charge
Support agent. Sends: the person's message
Finance decides. {{user.name}} asks: {{start.message}}
billing.refund
Finance couldn’t approve this one. A person will follow up today.
Done. I refunded {{charge.amount}} and emailed you the receipt.
Teams that run their AI through AiOrch
01See it run
Pick a run and watch it play: which model routed it, which tools the agent called, who approved what, and what it all cost. This is the trace your team gets for every single request.
Where’s my order #4821? It said Tuesday.
02Platform
Six building blocks that snap together. Start with one agent, grow into workflows, schedules and deploys without switching tools.
Give each agent a job, a model and a reading list. PDFs, sites and notes become searchable knowledge it cites back.
Routers, agents, tools, conditions and approvals on one canvas. Publish a version, roll back in a click.
Connect any HTTP API once. Share a key, or let each person sign in so agents act as them, never as root.
Mark a tool sensitive and every call waits in the right person’s inbox, with the exact arguments in front of them.
Point an automation at an agent or workflow, set a time, and get the result by webhook, email or feed.
Recipes run your deploy steps over SSH, watch your services, and can wait for an admin before touching prod.
03Guardrails
Every tool carries a sensitivity. Reads run, writes are capped, and anything that moves money, deletes data or ships code waits for the right person. Try the inbox.
Looks things up. Runs right away and lands in the trace.
Changes something. Runs within your rate limits and daily caps.
Moves money, deletes or ships. Waits for a named role to say yes.
Refund a duplicate March charge
Billing agent · asked by Sam Ortiz
04Developers
Drop a published agent on any site, call it from your backend, or use it as a tool from Claude Code, Cursor and VS Code over MCP.
<!-- One tag. No dependencies. -->
<script
src="https://aiorch.yourco.com/widget.js"
data-api="https://api.yourco.com"
data-agent="pub_support_8f2k"
data-theme="light"
async
></script>
05Usage
Spend, speed and volume by agent, person and day, with limits that stop a runaway loop before it reaches your invoice.
06Customers
We had four agents running in four different scripts. Now they live in one place, and I can see exactly what each one did and what it cost.
Per-person sign-in means the agent can only see what the person asking can see. Security signed off in a week.
Daily token caps per agent ended the surprise invoices.
The approval inbox sold it. Refunds over $100 wait for a person, everything else just happens.
Deploy recipes with an admin approval replaced a very scary Slack bot.
One script tag and our docs site had an assistant that cites the actual page it answered from.
Our ops team built the onboarding workflow themselves. Engineering reviewed the trace, not the code.
The morning pipeline report runs at 9:00 every weekday and lands in our channel. Nobody has touched it in two months.
When something goes wrong I open the run and see every model call and tool call in order, with timings. Debugging went from hours to minutes.
07Pricing
Everyone in your company can chat with your agents for free. You only pay for the people who build them.
For a team trying its first agents.
For teams putting AI into real work.
For companies with rules to follow.
Model usage is billed at cost with no markup · Cancel any time · Your data stays yours
08Questions
Pick a model per agent and per workflow step. Use a small fast model for routing and a stronger one for the hard parts; every call’s tokens and cost show up in the trace.
Yes. Connections can be shared, or set so each person signs in. Then an agent only ever sees and does what that person could do themselves.
You decide. Mark a tool, a workflow step or a deploy as needing approval, choose the role that decides, and the call waits in their inbox with the exact arguments.
Enterprise plans can self-host the whole platform. The web app only ever talks to your AiOrch API, and keys stay in HttpOnly cookies on your domain.
Publish the agent, add one script tag, and choose which sites may use it. Rate limits and daily message caps protect your bill.
Every agent has per-minute rate limits and daily message and token caps, and the usage page shows spend by agent, person and day.
Set up your first agent in ten minutes. Free for up to three builders, no card needed.