I’ve been running autonomous agents (OpenClaw/ReplyX) for lead gen, and I kept waking up to drained API credits because bots would get stuck in "zombie loops"—retrying a failed step (like a button click) until the timeout hit.

I actually wrote the fix for the loop detection logic in the OpenClaw repo itself (PR #9759), but I realized I needed a better way to visualize what was happening than scrolling through thousands of lines of JSON logs.

Empusa is a local dashboard (Next.js + SQLite) that helps with this:

Visual Timeline: Turns logs into a visual trace (Green=Success, Red=Fail, Yellow=Loop Detected).

State Snapshots: It records the agent's variable state at every step.

Resume Capability: If an agent crashes, you can grab the state JSON and resume execution from that exact step rather than restarting from zero.

It’s a simple MVP right now, but it solves the specific pain of "blind" agent debugging. Code is MIT licensed.

Would love feedback on the state management approach.