Stop re-explaining yourself
README
Stop re-explaining yourself.
You have a repo. You have a goal. You also have four CLIs open in four windows, each holding a slightly different version of the same task. By evening you've spent more time pasting the goal back into the next chat than actually building.
Goodboy is a desktop app that holds the goal, the plan and the context once, then hands them to whichever agent you want to run next. Same brief, same memory, different model. Conversation, plans, decisions and PR state stay in a local SQLite on your machine. Your keys, your data, your bandwidth.
<img width="3972" height="2234" alt="CleanShot 2026-06-10 at 10 38 18@2x" src="https://github.com/user-attachments/assets/0d9d8bf4-df65-4b48-97af-ca97f5f56109" />Switching between Claude, Codex, Cursor and Gemini ten times a day was
eating my afternoons. Every new tab meant rebuilding the same mental
scratchpad from scratch, then watching the next model run off in a slightly
different direction. Eventually I got fed up and built this.
Open source. Every feature included. No paywall, no telemetry, no account.
Shared context, not vendor sessions. Goal, decisions, last summary, open questions. A summarizer keeps it fresh after every turn. Edit any field by hand when the agents get it wrong. The next agent shows up already briefed.
Provider swap mid-task, without amnesia. Each turn is rebuilt from the shared context, never resumed from a vendor's session blob. Drop Claude halfway, hand the same task to Cursor, Codex or Gemini, watch it pick up clean.
Workflows for the multi-step stuff. Refactor incoming? Line up a sequence: a cheap model to scout the area, a smart one to plan it, a mid one to implement, another to review, a cheap one to open the PR. Each step picks its own provider and model, so you're never paying Opus prices to run a grep.
Plans as artifacts, not transcript scrollback. Agents write the plan before they touch your