Observability for assistant runs

Why this mattered

When an assistant gives a bad answer, the question is not only what it said but what context and tools shaped it.

This belongs in the development timeline because Agentic Assistants is not a single feature. It is a local-first assistant framework with a CLI, FastAPI and WebSocket server, MCP bridge, Next.js control panel, indexing, scoped retrieval, knowledge bases, pipelines, discovery, and training workflows. The project only became useful once its infrastructure decisions were written down well enough to be repeated.

Design decision

Telemetry and experiment tracking make sessions inspectable, especially when retrieval, model routing, and pipelines are involved.

The practical stack around this decision includes Python, Poetry, Click, FastAPI, WebSockets, MCP, LanceDB, Chroma, DuckDB, Polars, PyArrow, CrewAI, LangChain, LangGraph, Ollama, MLflow, OpenTelemetry, Next.js, Docusaurus. I try to keep the interfaces small: configuration describes intent, runtime code owns behavior, and operational notes explain what a future maintainer should check first.

What I would repeat

This is how local-first can still be accountable.

The repeatable pattern is to make the boring path explicit. For this project that means clear repository boundaries, documented setup, predictable deployment commands, and enough observability to know whether the system is healthy or merely quiet.

Reader takeaway

If you are building something similar, start with the workflow you need to repeat every week. Then add only the platform pieces that make that workflow easier to recover, explain, and extend.