oboeapp.com → dif.sh
Oboe is now dif
If a link to oboeapp.com sent you here, you're in the right place. Oboe is now dif.
Oboe started as an A/B testing tool: a snippet on your site, a dashboard to build variants, results in charts. This year we pointed it at agent-driven experimentation. Agents draft the tests, people approve what ships, and every result feeds the next idea. We built that as a dashboard, with a Launcher for approvals, a Pulse for what was running, and a History so learnings stuck around.
While building it, we kept hitting the same wall. Every piece we added to the dashboard
already existed in the repo. Approval was a pull request. History was git log.
And the context an agent needed to draft a good test sat in a dashboard it couldn't read,
one tab away from the code it could.
So we moved the experiments into the repo.
What dif is
dif puts feature flags, A/B tests, holdouts, and staged rollouts in your repo as Markdown files. One file per experiment, reviewed in the same pull request as the code it gates.
dif/surfaces/keeps a learning log for each part of your product.dif newreads the last three learnings into every new draft.dif/context.jsonis the first file your coding agent reads, so it knows what's running before it touches anything.dif buildcompiles the active files intodif/generated/, a typed client your app imports. Variants resolve locally, with no network call.- Eight CLI verbs and a small JS SDK. One command to install, no signup.
- dif Cloud adds a live pulse and a metrics catalog, and proposes changes back as pull requests.
What carried over
The ideas from Oboe are all still here. They're files now.
| In Oboe | In dif |
|---|---|
| Launcher: approve what goes live | Pull request review |
| Pulse: what's running, and how it's doing | dif Cloud live pulse |
| History: keep the learning | dif/surfaces/*.md and git history |
| Ideation: turn patterns into hypotheses | dif new drafting from the surface log |
The bet is the same one Oboe made: the teams that do well with agents won't be the ones that ship the most, they'll be the ones that learn fastest and keep control of what goes live. We just think that work belongs next to the code.
Try it
Install the CLI and run dif init in a repo. The first experiment guide walks you from an empty repo
to a running test.
Thanks for following along,
Chris Fowles