Now · v0.1
Local comparison and human evaluation
Run coding agents locally, inspect blind candidates, and record the outcome you prefer.
Dispatch Cloud · Future direction
Dispatch works without a Cloud account.
Evaluation sync is optional and requires explicit consent. Automatic routing and hosted execution are not available in v0.1.
Optional today
Opt-in evaluation sync
Preview eligible evaluation data, enable consent, and sync only with a configured token.
Future direction
Routing informed by real evaluations
Future versions may use opt-in evaluations from real tasks to help choose an agent for a task.
Why compare on your own work
Compare agents on work from your own repository.
Public benchmark scores leave out your codebase, constraints, and preferences. Dispatch lets you inspect each result and choose the one that fits.
How evaluations could inform routing
01
Local run evidence
Each run records candidate results and your evaluation.
02
Optional evaluation sync
You choose whether to send eligible evaluation data to Dispatch Cloud.
03
You choose the agents
In v0.1, every run names the agents to use.
Try Dispatch on a real task.
Run Codex CLI and Cursor Agent on the same task, compare their changes, and choose the result that fits your codebase.