Getting started · v0.1

Run your first Dispatch comparison.

Use one bounded task, two configured agent tools, and a repository you can safely restore.

Step 01

Install the v0.1 preview.

Prebuilt archives are the preferred path for macOS Apple Silicon and Linux x86_64. The download page includes checksum verification and PATH setup.

Choose your download View latest release

Install with Cargo

cargo install --git https://github.com/dispatch-ai-org/dispatch --locked
dispatch version

Source installation requires the current stable Rust toolchain and Git. For deeper CLI details, read the Core README.

Step 02

Configure Codex CLI and Cursor Agent.

v0.1 release testing covers Codex CLI and Cursor Agent. Install each tool, authenticate it with its own provider, and review the permissions it will receive.

Codex CLI

Installed, authenticated, and callable from your shell.

Cursor Agent

Installed, authenticated, and callable from your shell.

Step 03

Prepare a repository you can restore.

Begin with a clean, committed test repository. Initialize Dispatch, then run the doctor command to catch missing configuration before an agent starts.

Initialize and check

dispatch init ./my-project
dispatch doctor ./my-project

Step 04

Run the same task through both agents.

Describe one bounded task and name both agents. Because local agents can change files and run commands, Dispatch requires the unsafe-local acknowledgement.

Run candidates

dispatch run ./my-project \
  --task "Fix the retry race in the worker." \
  --harnesses codex,cursor \
  --backend local \
  --timeout 300 \
  --max-parallel 2 \
  --allow-unsafe-local

--allow-unsafe-local confirms that you accept the risks of local execution. Dispatch provides no security guarantee or sandbox.

Step 05

Inspect the saved run evidence.

Dispatch gives each candidate a blind label such as A or B. Compare the outcomes, inspect individual diffs, and review the retained run record before identities are revealed.

Inspect

dispatch compare <run-id>
dispatch diff <run-id> A
dispatch diff <run-id> B --stat
dispatch inspect <run-id> A
dispatch show <run-id>

Step 06

Record the result you prefer.

Choose a candidate and preserve why it fit the task. Evaluation is part of the compare command; there is no separate evaluate command.

Record an evaluation

dispatch compare <run-id> \
  --winner B \
  --reason correctness \
  --reason cleaner-change \
  --explanation "B fits the existing design better."

Apply only after review.

When you are satisfied, apply the selected candidate with:

dispatch apply <run-id> B

Current v0.1 limits.

The preview is local-first and CLI-driven. Automatic routing, hosted execution, billing, and a polished cloud dashboard are not part of the current product.

Review trust boundaries