Make your coding agent better.

Excellent runs next to the coding agent and model you choose. It checks the work against your repo and tools, learns from failures, and tries changes that make the agent more reliable on your codebase.

Agent sessioninvoice-app / checkout-validation
running locally
YouFix checkout validation without breaking signup.
Coding agentDone. I updated checkout validation and added tests.
ExcellentI'll check that claim against the repo and connected tools.
Evidence checksIndependent of the agent's claim
3 pass / 1 fail
DiffCheckout and test files changed
TypesNo contract drift
TestsUnit suite passes
BrowserSignup flow loses the selected plan
CIWaiting on corrected attemptHeld
DeployNo release observedHeld
What Excellent learnsFrom repeated, evidenced failures
Failure pattern

UI checks happen after too many changes are bundled together.

Candidate change

Run the browser flow after each checkout or signup edit.

Compare on held-out work before approval.
Decision supportAction follows evidence, not confidence
Return this attemptSend the browser failure back to the agent.
Recommend approvalOnly when a candidate holds up.
Keep current or roll backWhen evidence weakens after a change.

Use the coding agent and model you choose.

Install ExcellentView on GitHub
curl -fsSL https://excellent.so/install.sh | sh
cd your-repo
excellent init

The agent does not decide when it is done.

Your coding agent can say it finished. Excellent checks the result separately.

It can inspect the exact git tree, run the required tests, open the browser, check the build, and observe the deployment. When something fails, it sends the exact failure back to the agent. When it cannot tell, it says so.

Task
Add CSV export to the invoices table.

Coding agent
Done.

Excellent
Returned.

The export ignores the active invoice filter.
Expected 12 rows. Found 47.

Second attempt
Verified.
See how a task is checked

Then it tries to improve the setup.

After enough work, Excellent compares different instructions, models, tools, context, checks, and retry rules.

It tests the candidates on tasks the system proposing the changes did not see. Nothing is promoted because it merely looks better.

CANDIDATE COMPARISON

Quality       Held-out verified tasks
Cost          Cost per verified result
Help          Human intervention per task
Safety        Hard regressions
Decision      Recommend approval or keep current

A candidate is compared with the current setup across quality, cost, human help, and regressions.

CHECKING
- Run browser checks only at the end
+ Run browser checks after each UI change

RETRYING
- Retry the same command three times
+ Identify the failure before trying again

MODEL USE
- Use the largest model for the whole task
+ Use a smaller model for repo reading and test repair

Keep using the agent you already use.

Excellent is not another coding agent.

Your agent does the work. Excellent starts the attempt, records what happened, checks the result, and compares different ways of running it.

Excellent
   starts and observes
Coding agent
   changes
Your repo
   checked by
Git - tests - browser - CI - deployment

Runs on your machine.

Excellent installs a local service, a CLI, and a browser interface. Your repo and work history do not need to be uploaded to Excellent. Your coding agent still uses the model provider you choose.

  • Local database
  • Local artifact store
  • No account required for local use
  • Receipts can be checked offline
  • Uninstall and export are documented

You get the evidence, not just a score.

For any task, you can see:

  • What was requested
  • What the agent changed
  • What the agent claimed
  • What Excellent checked
  • What passed
  • What failed
  • What could not be checked
  • What happened after the change landed

We are building Excellent with Excellent.

We have not published a verified improvement yet. The current build status is available in the changelog.

These are our results on our repository. They are not a promise about yours.

Open the changelog

Try it on a real repo.

Run Excellent locally. See what your agent gets right, what it misses, and whether a better setup exists.

curl -fsSL https://excellent.so/install.sh | sh
cd your-repo
excellent init

Using coding agents across a team? Email Matt.