Extending VectorStep

VectorStep does not accept external contributions — but it is Apache-2.0, so extending your own copy is expected and supported (see Licence & contributions). These are the three extension points designed to need no other changes: source parsers, executors, and library steps. This page is the complete reference for all three — condensed pointers to it also appear on Webhooks, Executors, and Steps.

Adding a new source parser

1. Create src/normaliser/<source>.py
2. Implement BaseParser — produce a NormalisedContext
3. Register in src/normaliser/__init__.py source map

Adding a new executor

1. Create src/executors/<name>.py
2. Implement BaseExecutor — accept StepConfig + context dict, return LLMOutput
3. Register in src/executors/__init__.py executor map
4. Reference by name in pipeline YAML step executor: field

No other changes required in either case.

Adding a library step

1. Create steps/<your-step-name>.yaml with at minimum name, executor, and executor_config.agent
2. Run POST /reload (or send SIGHUP) — the step will appear in /ui/steps immediately
3. Reference it in any pipeline with - use: <your-step-name>

The steps/ directory is gitignored — steps are personal to your deployment. Copy starter definitions from samples/steps/ and adapt them, or write your own. See samples/pipelines/alert-triage-investigation-using-steps.yaml for a worked example of a pipeline that uses library steps.