AI code reviews that admit they aren't human yet
Sourcery auto-reviews GitHub PRs with OpenAI and Anthropic, but concedes it still doesn't match a colleague's eye.

What it does Sourcery is a GitHub app that drops AI-generated reviews on pull requests: change summaries, high-level feedback, and line-by-line comments. It also runs as an IDE assistant for Python-heavy workflows in VS Code, PyCharm, Sublime, and even Vim. Public repos get it free; private ones hit a 14-day wall then need a paid plan.
The interesting bit The README’s honesty is the standout feature. Sourcery explicitly states its goal is “the type of code review you would expect to get from a colleague” and admits “we’re not there yet.” That’s rare in the AI tools space, where most vendors pretend they’ve solved software engineering.
Key highlights
- Uses both OpenAI and Anthropic LLMs, not a single model bet
- Code retention capped at 30 days; no training on customer code (stated policy)
- IDE mode handles test generation, docstrings, and readability tweaks
- Free tier actually exists for open source, unlike many AI devtools
- Supports Vim, which suggests someone on the team has battle scars
Caveats
- The README is vague on what “pieces” each LLM handles and how prompts are structured
- “More editor integrations coming soon” has been a standard promise since 2019 in this space
- Pricing for private repos requires clicking out to docs; no numbers in-repo
Verdict Worth a spin for open-source maintainers drowning in PR volume. Teams with strict data residency requirements should verify that 30-day claim directly with Sourcery before feeding it proprietary code.