Loading...
🔒 Private Preview:Whet is currently in private preview. Features may change and some functionality may be limited.Join the discussion →
Loading...
Whet runs automated pull-request generation, review, and merge-readiness workflows on Temporal — so a network drop or process restart never loses hours of in-flight agent work.
Every stage is a durable Temporal activity — any crash or network drop is transparent.
GitHub App receives pull_request.opened or pull_request.synchronize and dispatches to the Whet workflow engine.
A durable workflow execution is created. From this point forward, any crash or network interruption is fully recoverable.
CrewAI agents run architecture review, security scanning, and diff analysis in parallel — with optional Sluice routing for failover and cost tracking.
Security thresholds, coverage minimums, and custom gates evaluate the crew's output. A failure blocks merge readiness.
Findings are filed as GitHub issues. Structured events (analysis.completed, issues.created) are emitted to configured notification channels.
PR is marked ready for maintainer merge. Status can be reported to a task tracker (e.g. Baton) if configured.
Multi-hour pipelines survive network drops and process restarts. Workflows resume exactly where they left off — no lost work, no full restart.
Triggers on pull_request.opened and pull_request.synchronize. Emits structured events: analysis.completed, issues.created, security.alert, quality.gate.failed.
Per-repo config controls AI provider, model, workflow triggers, quality gate thresholds, and notification channels. No code changes required.
Automated refactors run in ephemeral, isolated worktrees — clean commits guaranteed, main working tree never contaminated.
Test coverage, security scans, and gate approvals are verified before signaling PRs as ready — not just a diff someone has to eyeball.
Sluice can handle model routing and failover (optional). Baton can receive PR status as evidence receipts when configured. Retort quality gates can trigger workflow dispatch.
Whet is a runtime execution layer — a distinct role from the other tools it coordinates with.
When configured, Whet can report PR creation, review outcomes, and merge status back to Baton as structured evidence receipts on the originating task.
Outputs that pass Retort's deterministic quality gates can trigger Whet Temporal workflows for automated branch creation and PR synthesis.
Whet can route LLM calls through Sluice for failover, rate limits, and cost attribution — or call providers directly if Sluice is not deployed.
Whet AI spend is tracked per PR analysis run via Docket when Sluice is in the chain — so you know the cost of each automated review cycle.