Forked – a local time-travel debugger for OpenClaw agents

Name: Forked – a local time-travel debugger for OpenClaw agents
Availability: InStock
Author: TedOS

by TedOS·Feb 20, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●●BangerSolve My ProblemWizardryNiche Gem

OpenClaw time-travel debugger with fork-and-replay when logs are useless.

Strengths

•Solves a real pain: agent debugging without logs or visibility into decision points.
•Fork-and-replay from any decision point with payload modification is genuinely clever.
•Local-first, privacy-preserving architecture with no cloud or telemetry required.

Weaknesses

•Tightly coupled to OpenClaw; limited to single-agent framework, excludes other platforms.
•Early v1 product; unclear stability, adoption, and whether file snapshots scale to large writes.

Post Description

I built Forked because debugging OpenClaw agent failures was too opaque: lots of logs, hard to reproduce, and no clear way to inspect where behaviour diverged.

Forked is a time-travel debugger for OpenClaw that records agent execution and lets you inspect, fork, and rewind runs.

What it captures:

- LLM inputs/outputs (including token usage when available) - tool calls/results - message flow - file-modifying actions (with snapshots) - lifecycle/config/setup-file changes

What you can do:

- inspect a full timeline of a run - fork from a decision point, modify payload, replay - rewind file state to a prior snapshot-backed point, then branch from there

Architecture is simple/local:

OpenClaw tracer plugin -> WebSocket local daemon (Node + SQLite) local UI (React)

Privacy model:

MIT licensed local-first (127.0.0.1) no cloud dependency, no telemetry required Repo: https://github.com/MurbotLabs/Forked

This is v1. I’d really value feedback on:

- Missing debugging workflows - Pain points in fork/rewind UX - What would make this essential in daily OpenClaw dev loops