Deploy
Choose intelligence appropriate for the job.
Classifying an invoice status and deciding whether a customer's complaint is reasonable need different amounts of thinking. Agents are configured with the kind of model a task needs rather than being built around one provider, and the platform is not architected so that changing that is a rewrite.
Four choices, described by what they are for
A recommended default, a fast one for classification, a reasoning one for difficult decisions, and your own if you would rather supply it. Not a list of version numbers you have to track.
Changing the model is a version
It goes through review and is recorded on every subsequent action. A model change can alter behaviour as much as a permission change, and it should leave the same trail.
One agent can use more than one
Cheap classification, careful reasoning on the hard cases, document reading where needed. An automatic mode makes that choice per step, and you can pin it if you would rather decide.
Provider choice is not your cost problem
Which provider handles which task is ours to manage. You see the cost of running an agent, not a per-token bill you have to model in advance.
Agent Runtime
The rest of this area.
Isolated execution, resource limits, retries and durable state.
How agents execute
Isolated environments, resource limits, clean starts.
Retries and durable state
What happens when something fails mid-job.
Environments
Development, staging, production. Promote, don't rebuild.
Budgets and limits
Ceilings on actions, spend and calls. Hit one and it stops.
Agent teams
An agent can ask a specialist. The chain stays visible.
Give AI a job.
Build an agent, choose what it can access, define what it may do, and deploy it to work safely across your business.
No code required · Full SDKs and API · Stop any agent instantly · [email protected]