| name | wire-component-into-pipeline |
| description | Step-by-step workflow for integrating a new component (classifier, reranker, etc.) into an existing app pipeline with metrics and hooks |
/wire-component-into-pipeline
When to use
Use this when adding a new processing component (e.g. intent classifier, reranker, embedder) to an existing pipeline — requiring code edits across multiple files, metrics registration, and hook-in at the right execution stage.
Steps
- Implement the component — write or import the core module (e.g.
memory/reranker.py, classifiers/intent.py)
- Hook into the relevant data layer — modify the search/retrieval function (e.g.
memory/facts.py search()) to call the component
- Wire into the app entry point — import the component in
app.py, instantiate it, and insert it into the pre-LLM / pipeline path
- Register metrics — find the
_METRICS dict (or equivalent) and add counters/histograms for the new component
- Locate the correct pipeline stage — read the handler function (e.g.
on_user_turn_handler) to find exactly where to invoke the component relative to the LLM call
- Test end-to-end — verify the component runs in a real request and metrics increment correctly
Examples
- Adding an intent classifier: implement class → hook before LLM call in
app.py → add intent_classified_total to _METRICS
- Adding a reranker: implement rerank function → call inside
facts.py search() after vector retrieval → add rerank_latency_seconds histogram to metrics
