Turn unused Claude or Codex quota into reviewable PRs before the weekly reset
RemainderBot spends leftover AI subscription quota on finished pull requests in your own private repo, with merge or close as feedback.
- LinkLoot access
- Free
- Provider costs
- Unknown
What you get from it
What it does
RemainderBot watches the weekly quota clock on a Claude Code or Codex CLI subscription. When an hourly cron job sees that the reset is less than four hours away and at least 10% of quota remains, it kicks off a run. The agent reads task lists, issues, docs, and notes from configured sources plus a GOALS.md file you write, picks one unfinished task, commits a plan, then builds against a deadline sized to the remaining quota. Output is a document, a tested feature, or a prototype, delivered as a pull request in a dedicated private GitHub repo. Merge counts as approval; closing with a comment feeds the next run's selection logic.
Who it helps
Developers who pay for Claude or Codex subscriptions but regularly leave quota unused at week's end. It suits solo builders or small teams that keep project state in plain text files and want finished work rather than chat transcripts. Runs land in your own repo, so nothing touches shared infrastructure without your review.
Getting started
You need a dedicated machine such as a small VPS, Python 3.11 or newer, git, and the GitHub CLI (gh). Install either Claude Code or the Codex CLI (or both), logged in with an active subscription. Clone the repository directly rather than forking: forks of public repos cannot be made private, and template copies create merge conflicts on updates. Create a private repo from the clone, copy sources.example.toml to sources.toml, edit GOALS.md and the source config, then commit and push. The bot generates its own inbox and state files. If you use Google Workspace sources, install the gws CLI separately.
Limits and costs
The software itself is MIT-licensed open source, but running it requires a paid Claude or Codex subscription and hosting on a machine you control. Every instance must stay private because each run branch embeds a copy of everything the agent read along with its full log. The design runs the agent with permission checks disabled on the dedicated box, treating that machine as the sandbox. Finalize passes can only cut scope, never add features. Pricing beyond the underlying subscription is not specified by the project documentation.
Source links
Discussion
Share practical experience, questions, or warnings with the community.
Sign in to join the discussion and vote on comments.
Sign in