Work on Repeat is out of private testing. Anyone can now create an account, connect the tools they already use, and put recurring knowledge work on a schedule.
The problem we set out to fix
Most valuable AI work is not a one-off prompt. It is the Monday pipeline review, the weekly security audit, the monthly analytics report — work that must happen again and again, the same way, with the same access, and with a record of what actually ran. Chat tools forget. Homegrown cron jobs grow into unowned infrastructure. Automation builders bury the work in flowcharts that nobody wants to maintain.
Schedule first, not chat first
A routine here starts from the schedule: instructions, validated inputs, a model, a timezone-aware cron expression, and execution limits. Runs are bounded by time, stream a live trace while they execute, and keep that trace afterward so you can see every model step, tool call, and result — not just the final output. The routine library ships ready-to-adapt starting points, and use cases describe the recurring responsibilities they cover.
Permissions you can read in one screen
Every discovered tool starts disabled. Enabling a tool is an explicit choice, and letting it run without a human in the loop is a second, separate choice. Connected providers run through reviewed, fail-closed policies — an unreviewed tool or a drifted schema simply does not reach the model. The integrations catalog shows what each connection can and cannot do, and the security page explains the boundaries in depth.
Pricing
Self-serve billing has four tiers: Free to prove a routine, Starter for a first paid routine, Pro for a team's working set, and Scale for many concurrent routines with full API access. The pricing page has the current comparison.
What it does not do yet
We would rather tell you now than surprise you later. Several integrations are connection-only previews while we review their tool schemas — connecting works, but tools stay disabled until the review lands. Runs are bounded by wall-clock limits, so open-ended crawls are deliberately out of scope. And there is no visual workflow builder; instructions are written, versioned prose.
Where to start
Pick a routine from the library, or start from your own recurring responsibility and sign up. Developers can create routines and receive signed result webhooks through the API. The changelog records what ships from here.