Skip to content
AdiosDocumentation
Browse documentation

Run agents on Adios

The Agent Builder lets a team create, configure, test, and operate agents. Start from a template, a description, or a blank agent. The underlying kit files remain editable.

Configure and test

  1. Open or create an agent and review its instructions and workflow.
  2. Set its configuration, bind the required connections, and choose approval modes.
  3. Save the agent. New agents start paused.
  4. Run a test and inspect Activity, step outputs, and generated files.
  5. Review any simulated actions before enabling normal operation.

Test runs use real model calls and read operations. Other connector effects are simulated; supported database tests use temporary copies. Approvals and delays are skipped and recorded. Tests can consume model and service usage.

Work from files

adios agent create --dir ./hello-workflow --name "Hello workflow"
adios agent show AGENT_ID
adios agent run AGENT_ID --test --workflow hello --input '{"name":"Adios"}'
adios agent logs RUN_ID
adios agent files RUN_ID --out ./run-files

Use the IDs returned by the previous commands. The quickstart provides these example files. Connectors use installation bindings; secret values stay outside the kit.

adios agent set AGENT_ID --bind CONNECTOR=CONNECTION_ID
adios agent set AGENT_ID --approval ACTION_CLASS=manual

Choose real connector and approval names declared by your kit. Keep manual review for actions you want to inspect before execution.

Versions and ongoing operation

Saving a valid agent creates and deploys a version. A draft can remain invalid while you edit it. Each run uses its saved version so later edits do not change that run's instructions halfway through execution.

adios agent pull AGENT_ID ./my-agent
adios agent validate ./my-agent
adios agent push AGENT_ID ./my-agent
adios agent versions AGENT_ID

Review the resulting version before resuming scheduled or event-triggered work. Use adios agent resume AGENT_ID when ready, and adios agent pause AGENT_ID to pause the agent. Inspect pending approvals and run errors in Activity.

Hosted function invocation has its own immutable kit qualification and execution path. It is currently separate from Agent Builder workflows, which reject function steps. See functions and testing before adding custom code.