Riven Computer
LiveHand off whole tasks instead of single questions. Background runs with live streaming output, persistent history, cancel and retry, and honest failure states.
- Service
- riven-computer
Chat answers questions; Computer takes *tasks*. Describe an outcome — "research what changed in this market this week and summarize it" — and Computer works it in the background, streaming progress and keeping every run in your history.
How it works#
- Describe the task. Write the goal in plain language. More context in, better results out.
- Watch it run. Output streams live as the task executes, rather than arriving as one final blob.
- Keep the result. Every run is saved with an explicit status: complete, failed, or cancelled.
Cancel and retry#
A running task can be cancelled at any point — it stops promptly and is marked cancelled, never silently discarded. Retrying a failed or cancelled task creates a fresh linked run, so the original record survives and attempts can be compared.
Honest by design#
Computer never dresses a failure up as a success. A failed task shows an explicit failure state plus whatever was produced before it failed. Each run records how it was metered, so usage is always attributable back to a task.
Rolling out#
The next generation of Computer is rolling out now: scheduled tasks (interval or cron), multi-agent runs where coordinated workers research and synthesize with citations, plan-first mode (approve the plan before anything runs), model council (several models answer independently and disagreements are surfaced), and reusable skills and workflows. Watch the changelog for availability.
When to use what#
| You want… | Use |
|---|---|
| An answer now, interactively | Riven Chat |
| A whole piece of work done in the background | Riven Computer |
| Riven inside your own product | The API |
| A repeatable evaluation of either | Riven Evals |