Launched this week

Wayfinder

Launched this week

Turn scattered AI work into experience you can reuse.

2 followers

Turn scattered AI work into experience you can reuse.

2 followers

Wayfinder turns AI coding sessions into a branching voyage map you can revisit and reuse. It records prompts, tool calls, file changes, tests, decisions, and wrong turns locally. Restore any waypoint without deleting the abandoned route, compare alternative approaches, and see which path actually worked. The first production adapters support TraeCode, Claude Code, and Codex through an IDE sidebar, in-conversation MCP App, and terminal view. Open source, no telemetry, no Wayfinder cloud.

Launch tags:Developer Tools•GitHub

Launch Team

LightfieldAI-native CRM that builds itself and does work for you

Promoted

Maker

📌

I built Wayfinder after noticing that AI tools were getting better at producing answers but not at preserving the reasoning that made those answers useful. Existing checkpoints answer “can I go back?” Wayfinder asks a different question: “which path worked, which path failed, and what should I reuse next time?” Each prompt becomes a waypoint, with tool calls, file changes, tests, and human judgment attached. When you restore an earlier state, Wayfinder keeps the abandoned route visible and starts a new branch. Version 0.3.3 is local-first, open source, and available for TraeCode, Claude Code, and Codex through GitHub Releases, Homebrew, Scoop, Agent Skills, and the official MCP Registry.

Report

2d ago

Forum Threads

p/wayfinder-5

•

2d ago

What part of your AI workflow do you wish you could revisit?

Most work with AI is scattered across one chat after another. Once a task is done, the code may remain, but the failed attempts, evidence, and reasoning often disappear.
I built Wayfinder to test a different model: keep prompts, tool activity, file changes, validation results, verdicts, and branches local, then turn them into a voyage map you can inspect and recover from without deleting abandoned paths.
The first production adapters focus on AI coding across TraeCode, Claude Code, and Codex. When you return to AI-assisted work days later, what context is hardest to reconstruct: why a decision was made, which attempt failed, or how to reuse the successful path?

View all