Inspiration -
The inspiration from Synqit came from the workflow of a typical hackathon. Teams are fast-paced, everyone pushes to main, and merge conflicts are inevitable unless carefully managed, which contradicts with the fast-paced theme itself.
What it does
Synqit turns a feature intent into a commit pushed straight to main. An LLM integrates it (scoped via the codebase's dependency graph) against the current baseline, retrying automatically if main moves first.
How we built it
Built entirely in Jac (Object-Spatial Programming): the codebase is modeled as a graph of files and dependencies, walkers traverse it to compute blast-radius context, and byLLM drives the integration; all wrapped in a local CLI that talks to git directly, with no server in the push path.
Challenges we ran into
A major challenge we ran into was gaining repository control straight from Github and implementing an AI agent to push and commit to github automatically.
Accomplishments that we're proud of
We're proud of an insight we had that every project repository was really just a persistent graph. This could be modeled with Jac, which let us spawn walkers to figure out what files to change when integrating diffs.
What we learned
We learned about Jac, and the uses of OSP and graph-native languages. It helped us reframe something as fundamental as a repository to be a graph.
What's next for Synqit
The next goal for Synqit is testing on genuinely large codebases and examining how it succeeds and fails.
Built With
- agentic
- ai
- claude
- graph
- jac
- osp
Log in or sign up for Devpost to join the conversation.