OpenCommit
We built OpenCommit to fix something small that costs teams big: messy, low-signal commit messages. It turns staged diffs into clear, conventional messages using LLMs directly in the developer flow: github.com/di-sukharev/opencommit.
Adoption followed because the product stays close to the command line: one command, consistent output, no ceremony. OpenCommit has 7,400+ GitHub stars, 300,000+ npm downloads, and roughly 2,500 new downloads a week.
For teams, the value is operational: cleaner audit trails, faster reviews, easier onboarding, and release notes that start from readable intent instead of scattered commit noise.
The technical decision was to keep the tool almost invisible. Developers keep git, hooks, conventional commits, GitMoji, commitlint, and local or hosted model providers; OpenCommit just removes the repetitive decision point.
AI-TDD
AI-TDD began as a proof of concept for a stricter AI coding loop: write a failing test, let an LLM iterate in a REPL, run the suite, and refine code until the test passes: github.com/di-sukharev/AI-TDD.
The project reached 270+ GitHub stars and early inbound from engineers and investors exploring AI-native development. People used it for katas, prototypes, and guarded experiments with AI pair-programming.
The important constraint is that tests stay the source of truth. The loop is inspectable, humans stay in control, and failure is visible instead of hidden behind a vague generated patch.
AI-TDD taught us where AI helps most: scaffolding, refactors, and small units. It also made the boundary clear: architecture, security, and domain nuance still need human judgment.
Developer tools
Both projects follow the same product principle: developer tools should sit inside the existing workflow, not drag the team into another dashboard.
That is the technical bar for our open source work: small surfaces, clear feedback loops, practical defaults, and automation that makes important engineering work more visible.