Knows your code
Ask how something works and get an answer grounded in the files and functions weft actually read, not a guess.
weft is a coding agent for your GitHub repositories. It reads your code, makes the change in an isolated sandbox, proves it against your tests, and opens a pull request for you to review.
The LRU cache evicts the wrong key after a read. Fix it.
get() read the entry but never moved it to the front, so a key you had just used could be evicted. It now re-inserts on read, and a test covers it.
Features
weft does the legwork, searching, editing and testing, inside guardrails it can’t step outside of.
Ask how something works and get an answer grounded in the files and functions weft actually read, not a guess.
Every chat runs in its own sandbox with no network and no credentials, so nothing it tries can reach your machine or your secrets.
Before a change counts as done, your tests run again in a fresh environment. Anything that looks like a key or a weakened test is stopped at the door.
Just want answers? Ask mode is read-only: weft can look at everything and change nothing.
Your chat keeps its workspace warm, so the next message picks up right where the last one left off.
Commit when you’re happy, with a message drafted for you. weft opens the pull request and keeps it up to date.
How it works
Point weft at a repository and say what you need, in plain words.
Every search, edit and command streams in as it happens.
Follow up, change course or ask why. It’s one conversation.
Your tests run again in a clean sandbox before anything is done.
Commit and open the pull request in two clicks.
Quality
Every change to weft’s agent runs against a suite of real tasks, from bug fixes and features to follow-ups and trick questions, graded by hidden tests rather than by another AI.
Read the latest results