Recently shipped

FinishIt

Give FinishIt a stale repo. It clones, diagnoses the smallest useful finish line, patches the code, runs checks, and opens a PR with the work explained.

Run it See workflow

one command

npx @finishit/finishit github.com/you/that-repo-from-2024

finishit runlocal first

Clone and indexRepo shape, package scripts, stale branches, and obvious breakpoints.

memory.db

DiagnosePick a tight task that can be finished without inventing a product roadmap.

smallest PR

Patch and testUse the configured model, run the project checks, and keep the diff focused.

green gate

Open the PRSummarize what changed, what passed, and what still needs a human.

ship

0dashboard required

5provider paths

1clean PR

no telemetry by defaultnexus memory

What it finishes

FinishIt is for repos that are close enough to rescue: half-built CLIs, abandoned demos, broken package upgrades, and apps that need one disciplined pass before review.

Repo triage

Finds package managers, tests, entrypoints, and the likely failure path before touching code.

finishit diagnose
scripts -> test, build
risk -> stale deps

Scoped patch

Asks the configured model for a narrow implementation and keeps generated work tied to the discovered task.

task -> make CLI usable
diff -> focused
mode -> PR-ready

PR handoff

Runs the checks it can run locally and writes the pull request summary around evidence, not vibes.

checks -> build/test
summary -> commands
handoff -> limitations

Run it with your model

Provider routing is controlled by the model name or by the tools already on your PATH. Use Codex or Claude Code for subscription-backed runs, APIs for hosted models, and Ollama for local work.

Ollama

Local, no API key. Any non-Claude and non-OpenAI model id routes to Ollama.

ollama serve
ollama pull qwen2.5-coder:14b
npx @finishit/finishit . --model qwen2.5-coder:14b

Anthropic

Use the Claude Code CLI or an Anthropic API key. Claude model names route to Anthropic.

npm i -g @anthropic-ai/claude-code
npx @finishit/finishit . --model claude-code

export ANTHROPIC_API_KEY=sk-ant-...
npx @finishit/finishit . --model claude-sonnet-4-6

OpenAI

Use Codex CLI or an OpenAI API key. GPT and o-series names route to OpenAI.

brew install codex
npx @finishit/finishit . --model codex

export OPENAI_API_KEY=sk-...
npx @finishit/finishit . --model gpt-5

built on nexus prime

Finish the repo, not the fantasy roadmap.

The tool biases toward boring, reviewable progress: local clone, narrow diagnosis, explicit provider selection, checks before PR, and no hidden dashboard dependency.

01
No provider lock-inCodex CLI, Claude Code, OpenAI, Anthropic, and Ollama are all first-class paths.
02
Local evidenceIt records what it ran and what failed so the PR handoff is reviewable.
03
Small diffsThe goal is a shippable improvement, not a speculative rewrite.