I got tired of Docker + Postgres just to debug a local AI agent, so I built this

Pooja Ramachandran — Mon, 17 Aug 2026 10:23:49 +0000

Every time I wanted to check token usage or cost on a local smolagents
run, I ended up spinning up a full observability stack for what should
be a two-second check. So I built AgentBox — single Python file, zero
dependencies, logs to local SQLite, live dashboard at localhost:8888.

Nothing leaves your machine — no signup, no cloud, no config. Drop
agentbox.py in your project, wrap your agent function with a decorator,
done.

Currently tuned for smolagents (auto-detects token/cost data from the
.monitor object), but the core tracing works with any Python function.
LangChain/CrewAI adapters are next.

Repo: https://github.com/pujaaaaaa/agentbox

Would love feedback — especially from anyone doing local agent work,
what's missing or what you'd want tracked.

DEV Community: Pooja Ramachandran

I got tired of Docker + Postgres just to debug a local AI agent, so I built this