Learn how an agent harness works
by building one in Go
hango is a multi-tenant agent harness built one working step at a time.
Every step ships a tested binary and a lesson explaining why it exists.
What a harness contains
The model is the smallest part. The roadmap walks through the layers around it: the agent loop, durable runs, middleware, tenancy and observability.
One lesson per phase
Each phase ends with a learning doc comparing hango’s choice against goclaw and langchain deepagents.
Observability first
OpenTelemetry from phase 1, feeding Grafana LGTM and Langfuse, so every run can be traced and debugged.
The build plan
Twenty phases from an empty Go module to a production-ready server, each with its contract, tests and effort.
Architecture diagrams
System architecture, the agent loop, the run sequence and the observability pipeline.
Build journal
Session notes on what was decided, what broke and what comes next.