Dyff reads the whole PR before you do. By the time you open it, the 34 files are grouped into 4 - 5 reading stages: the auth changes first, then the migration, then the frontend wiring. You read in dependency order instead of triaging a file tree.
Cmd+K opens symbol search across every file in the codebase, not just the files in the PR. Jump to a definition, find every caller, read a type, come back to exactly where you were. No clone, no branch switch.
You evaluate them while you already fully understand the code.
Still not sure? Chat with the LLM to trace that call-site for you
Self-host on your own infra. Nothing about the PR leaves your network unless you want it to.
13 providers supported. Dyff takes nothing on inference.
No. Dyff is read-only — it reads your repo, runs analysis, shows you the diff. It doesn't write code, open PRs, or push commits
You will ned a Github Personal Access Token (PAT) so Dyff can access your repo and PR
No, we don't collect any of your codes
Dyff supports 13 LLM providers. You can point it at any models. We recommend using frontier models for better results