Same task, same starting point
Dispatch gives each agent an equivalent copy of your repository, so candidates never inherit one another’s changes.
Experimental developer preview · v0.1
Compare agents on your code.Keep the evidence.
Dispatch runs the same software task through multiple coding agents. Each starts from equivalent code, so you can inspect both runs and choose the result you prefer.
Read the quick start~/retry-service
$dispatch run . --task "Fix retry behavior" --harnesses cursor,codex
Same task · same baseline
Cursor completed18.2s
Codex completed2m 31s
Evidence saved locally: run_01J7K4
$ dispatch compare run_01J7K4
Local-first
runs on your machine
CLI-driven
runs from your terminal
Codex + Cursor
tested in v0.1
Opt-in sync
cloud is optional
Why Dispatch
Your codebase is the benchmark that matters.
Coding agents differ in speed, completion behavior, and implementation choices. Public benchmarks cannot tell you which result you will prefer on the task in front of you.
Inspect the full run
Review each diff, runtime, log, and configured check result before you choose.
Choose the result yourself
Dispatch stores the candidate you picked and your reasons alongside the run.
Inspect before you decide
See the work behind each result.
The marketing demo names Cursor and Codex for clarity. In the CLI, comparison uses Candidate A/B labels until you record an evaluation.
- Candidate diffs, logs, and output
- Configured project-check results
- Runtime and change size
- Your choice and why you made it
Run run_01J7K4
Fix retry behavior
Candidate A
Completed- runtime
- 18.2s
- diff
- +42 −11
- evaluation
- not recorded
Candidate B
Completed- runtime
- 2m 31s
- diff
- +31 −8
- evaluation
- not recorded
You decide before identities are revealeddispatch compare
How it works
Run both candidates, compare their work, and record your choice.
01
Describe one bounded task
Point Dispatch at a repository and give each agent one specific job.
02
Start from equivalent code
Dispatch prepares an independent candidate state for every agent you select.
03
Inspect each result
Compare diffs, logs, runtimes, and project checks while agent identities stay hidden.
04
Record your choice
Select the candidate you prefer, explain why, and keep the evaluation on your machine.
Product truths
Runs and evaluations stay on your machine by default.
Dispatch stores snapshots, run evidence, and human evaluations locally. Evaluation sync stays off until you enable it.
Read the trust overviewLocal evidence
Runs, snapshots, diffs, and evaluations are stored on your machine by default.
Source stays out of Cloud sync
The evaluation payload excludes source, patches, logs, paths, and credentials.
Opt-in sync
Local runs and evaluations work while Cloud sync is off.
Preview software
v0.1 is an experimental CLI preview.
Dispatch coordinates third-party agent tools. Codex CLI and Cursor Agent may communicate with their own providers, so their terms and data controls still apply.
Quick start
Try it on a task you already know.
Start with a clean, committed repository you can restore. Configure Codex CLI and Cursor Agent, then give both the same bounded task.
First comparison
dispatch run . \
--task "Fix retry behavior" \
--harnesses cursor,codex \
--backend local \
--allow-unsafe-localReal coding agents can modify files and execute commands. Review their permissions and begin with a repository you can safely restore.
Future direction
Today’s evaluations could guide future routing.
v0.1 compares outcomes and saves your evaluations. Future versions may use that history to route tasks. Automatic routing and hosted execution are not available today.
Run one real comparison
See the difference on your own code.
Download the preview and run one bounded task through Codex CLI and Cursor Agent.