Doistbot runs 4,500 code reviews a month at under $2 each, with seven parallel review passes and dynamic model routing that cut costs 30%.
Doistbot, an automated code reviewer that runs roughly 4,500 reviews per month across Todoist's codebase. It started as a Gemini CLI call in a GitHub Action, evolved into a multi-model council that voted on findings, and now uses multi-focus reviews with dynamic routing to match the right model to each PR.
Each PR triggers up to seven parallel review passes, each focused on a specific area: general correctness, reuse, code quality, efficiency, tests, internal standards, and security. Dynamic routing classifies each PR into one of six categories, from full review down to skipped, so routine PRs like dependency updates or translation syncs get cheaper, faster treatment while substantive changes get full attention from more capable models. The model mix includes GPT-5.4, DeepSeek V4 Pro, GLM 5.2, and Gemini 3.6 Flash as a fallback.