Agents & models
Agents
Built-in agents and the two ways to build your own.
An agent is one step's brain. Premia ships built-in agents, and you can add your own.
Built-in agents
| Slug | Name | Role |
|---|---|---|
plan | Planner | Breaks a ticket into per-repo tasks. |
execute | Developer | Writes code and opens a PR. |
review | Reviewer | Code review. |
qa | QA | Runs tests/build. |
requirements | Requirements | Checks the PR solves the ticket. |
revise | Reviser | Applies feedback and loops back. |
rebase | Rebaser | Resolves branch conflicts. |
merge | Merger | Merges the PR. |
Opening a built-in agent shows its stage and model. Built-ins run on Premia's managed infrastructure and can't be edited — but Duplicate as custom agent seeds an editable copy from one.
Custom agents
Paid feature
Creating custom agents requires a paid plan.
Two kinds:
- Config agents — a prompt + model. No code; you configure the system prompt, model, and what context it receives. Good for custom review lenses or formatters.
- Code agents — a Python
run(context)function. Full logic, deployed to our cloud. Premia hosts the infrastructure; you just writerun(). See Custom code agents.
Both plug into any pipeline step and receive the context bindings you choose (the run, repo, PR diff, plan output, secrets, etc.).