> ## Documentation Index
> Fetch the complete documentation index at: https://docs.qredence.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# fleet-rlm CLI reference: optimize, chat, serve-api

> Reference for the fleet-rlm and fleet commands: serve-api, GEPA module and skill optimization, terminal chat, Daytona smoke, and Web UI startup.

fleet-rlm exposes two command entrypoints:

* **`fleet-rlm`** — Typer-based command group with subcommands for server modes, optimization, and terminal chat.
* **`fleet`** — Lightweight launcher for terminal chat and Web UI startup.

## `fleet-rlm`

```text theme={null}
Usage: fleet-rlm [OPTIONS] COMMAND [ARGS]...

Commands:
  serve-api      Run the FastAPI server surface (used by `fleet web`).
  optimize       Run DSPy optimization workflows.
  chat           Start standalone in-process interactive terminal chat.
  daytona-smoke  Run a native Daytona smoke validation without invoking an LM.
```

### `fleet-rlm serve-api`

Run the FastAPI HTTP/WebSocket server. This is the backend for the Web UI and is invoked by `fleet web`.

```text theme={null}
Usage: fleet-rlm serve-api [OPTIONS]

Options:
  --host TEXT      Bind host          [default: 127.0.0.1]
  --port INTEGER   Bind port          [default: 8000]
```

**Examples:**

```bash theme={null}
# Default host:port
uv run fleet-rlm serve-api

# Bind to all interfaces on a custom port
uv run fleet-rlm serve-api --host 0.0.0.0 --port 8080
```

### `fleet-rlm chat`

Start standalone in-process interactive terminal chat with the RLM agent.

```text theme={null}
Usage: fleet-rlm chat [OPTIONS]

Options:
  --docs-path PATH         Optional document path to preload as active context
  --trace / --no-trace     Enable verbose thought/status display
  --trace-mode TEXT        Trace display mode: compact, verbose, or off
  --volume-name TEXT       Optional Daytona volume name for persistent storage
```

**Trace modes:**

| Mode      | Description                      |
| --------- | -------------------------------- |
| `compact` | Condensed trace output (default) |
| `verbose` | Full thought/status display      |
| `off`     | Disable trace output             |

**Examples:**

```bash theme={null}
# Interactive chat with default trace
uv run fleet-rlm chat

# Preload a document as active context
uv run fleet-rlm chat --docs-path ./docs/architecture.md

# Verbose trace output
uv run fleet-rlm chat --trace --trace-mode verbose

# Persist chat state to a specific Daytona volume
uv run fleet-rlm chat --volume-name my-volume
```

### `fleet-rlm optimize`

Run offline GEPA optimization for a registered DSPy module or a Fleet skill markdown artifact.

```text theme={null}
Usage: fleet-rlm optimize [OPTIONS] MODULE [DATASET]

Arguments:
  MODULE    Registered module slug (use 'list' to see available modules).
  DATASET   Path to JSON or JSONL dataset.

Options:
  --output-path, -o PATH      Where to save the optimized DSPy module artifact.
  --train-ratio FLOAT         Training split ratio for compilation. [default: 0.8]
  --auto TEXT                 Optimization intensity: light, medium, heavy. [default: light]
  --skill-name TEXT           Optimize a bundled or mounted Fleet skill instead of a module.
  --skill-path PATH           Optimize a SKILL.md-compatible markdown file.
  --trace-bundle-path TEXT    Offline trace bundle path for the RLM-GEPA proposer (repeatable).
  --report                    Print a markdown summary after optimization.
```

**Examples:**

```bash theme={null}
# List registered optimization modules
uv run fleet-rlm optimize list

# Default light GEPA optimization
uv run fleet-rlm optimize my-module dataset.jsonl

# Heavy optimization with a custom output path
uv run fleet-rlm optimize my-module dataset.jsonl \
  --auto heavy \
  --output-path ./artifacts/optimized.pkl

# Optimize a bundled skill with a distilled trace bundle as proposer context
uv run fleet-rlm optimize skill data.jsonl \
  --skill-name optimization \
  --trace-bundle-path artifacts/traces/optimization-failures.jsonl \
  --auto medium \
  --report

# Generate a markdown report
uv run fleet-rlm optimize my-module dataset.jsonl --report
```

<Note>
  Skill optimization runs the RLM-GEPA instruction proposer inside an isolated sandbox and requires `DAYTONA_API_KEY` to be configured. Registered module optimization runs locally and has no Daytona requirement.
</Note>

<Note>
  Optimized artifacts are saved for manual review and are **not** auto-loaded into the live runtime. Promote artifacts deliberately through the Optimization page or the `/api/v1/optimization/runs/{run_id}/promotion-drafts` endpoint.
</Note>

### `fleet-rlm daytona-smoke`

Run a native Daytona smoke validation without invoking an LM. Useful for verifying credentials and sandbox lifecycle in isolation.

```text theme={null}
Usage: fleet-rlm daytona-smoke [OPTIONS]

Options:
  --repo TEXT   Repository URL to clone into the Daytona sandbox.  [required]
  --ref TEXT    Optional branch or commit SHA to checkout after clone.
```

**Example:**

```bash theme={null}
uv run fleet-rlm daytona-smoke \
  --repo https://github.com/Qredence/fleet-rlm.git \
  --ref main
```

## `fleet`

Lightweight launcher for terminal chat and Web UI startup.

```text theme={null}
usage: fleet [-h] [--docs-path DOCS_PATH]
             [--trace-mode {compact,verbose,off}]
             [--volume-name VOLUME_NAME]
             [{web}]
```

**Examples:**

```bash theme={null}
# Terminal chat
uv run fleet

# Web UI (alias for `fleet-rlm serve-api` + SPA)
uv run fleet web

# Preload a document into terminal chat
uv run fleet --docs-path README.md

# Verbose terminal trace
uv run fleet --trace-mode verbose
```

Hydra overrides are supported as `key=value` tokens after the command.

## See also

* [HTTP API reference](/fleet-rlm/reference/http-api)
* [Configuration reference](/fleet-rlm/reference/configuration)