Optimize agents

Optimize is a local prototype. Demos are by arrangement. There is no hosted optimizer, public API or install command.

What a demo covers

1. Find repeated work in your traces and source.
2. Propose one small change.
3. Compare a baseline and a candidate on representative tasks. Count model calls and checking overhead.
4. A person reviews the diff, results and limits, and decides whether to adopt it. If the evidence does not support the change, keep the baseline.

Bring to a demo

• A redacted trace
• The relevant source
• Representative tasks
• Quality criteria, and how you measure cost

In a demo, you choose what source or traces to share and whether to apply any change.

What the example shows

The example on the Optimize page is synthetic. A baseline made 2 catalog dispatches and the candidate made 1. These are work counts, not tokens or dollars. The quality result is a toy and does not show quality on your work. Checking overhead prevented adoption.

The workflow demo at https://slancha.ai/demo/workflow-review.html is illustrative. It stays in the page: nothing runs or deploys, it shows no real run, and it shows no general optimization result.

The local prototype runs one bounded synthetic job: a narrow registered repair that prepares an immutable catalog once per trial instead of twice. It does not optimize arbitrary uploaded code. It has no real provider transport. It does not show savings or a general result. A reference run keeps the baseline, because the overhead of checking outweighed the saving.

What is not available

• No hosted optimizer or public API.
• No upload of your code for automatic optimization.
• No customer results or savings claims.
• Automated efficiency discovery is not a released hosted feature.

Start with an audit

The free audit shows recorded patterns in one trace. It does not change code or prove an improvement.

To request a demo, use https://slancha.ai/contact.