> For clean Markdown content of this page, append .md to this URL. For the complete documentation index, see https://docs.nvidia.com/dynamo/llms.txt. For full content including API reference and SDK examples, see https://docs.nvidia.com/dynamo/llms-full.txt.

# AI Simulate (Experimental)

<Warning>
**Experimental.** AI Simulate is intended for evaluation and feedback, not production capacity
planning. Its Python APIs, configuration schemas, search results, and deployment output may
change without a standard deprecation period.
</Warning>

AI Simulate is a standalone Python distribution. It provides inference-engine forward-pass
simulation, deployment simulation, and search without depending on `ai-dynamo`.

## Sweeper

[Sweeper](/dynamo/dev/knowledge-base/modular-components/ai-simulate/sweeper/overview) searches backend deployment settings against an injected replay runner.
Its core owns backend search, candidate orchestration, scoring, and the versioned `ReplaySpec`
contract.

Optional adapters extend the search without adding a Dynamo dependency to AI Simulate. The
`ai-dynamo` wheel registers the `dynamo.planner` and `dynamo.router` adapters. Selecting either
adapter imports its Dynamo implementation and adds a versioned runtime hook to the replay
specification.

KVBM search settings are deprecated and have no adapter migration. Native G2 replaces that path.

## Install

The `dynamo-planner` image builds and installs AI Simulate and Dynamo from the same source revision.
The AI Simulate wheel remains inside the image and is not published as a standalone release
artifact.

For source development, install AI Simulate, Dynamo, and the Planner dependencies from the
repository root:

```bash
uv pip install -e ./aisimulate
uv pip install --no-deps -e .
uv pip install -r container/deps/requirements.planner.txt
```