# CLI And Server Surface

## CLI

Recommended shipped commands:

```text
nexum doctor --model /models/Nexum
nexum chat --model /models/Nexum --prompt "Summarize this workspace."
nexum agent --model /models/Nexum --prompt "Inspect this workspace." --workspace /workspace
nexum tools list
nexum tools exec-smoke
nexum templates list
nexum serve --model /models/Nexum
nexum status self-correction
nexum status self-improvement
nexum outcome record --model /models/Nexum --session-id SESSION --turn-id TURN --score SCORE --evidence-sha256 DIGEST
```

## Server

OpenAI-compatible endpoints:

```text
GET  /health
GET  /v1/models
POST /v1/chat/completions
POST /v1/responses
```

Nexum-native endpoints:

```text
GET  /tools
POST /agent/run
POST /tools/execute
POST /outcomes/record
GET  /self-correction/status
GET  /self-improvement/status
GET  /templates
POST /model/validate
```

Server behavior:

- validate and load the full Nexum folder before binding the socket
- keep model files read-only
- keep session model state under `NEXUM_STATE_DIR`
- keep mutable workspace tools and action receipts under `.nexum/`
- expose tool schemas
- persist observations
- return correction receipts
- redact secrets from logs
- require bearer authentication for non-loopback binding
- keep HTTP tool execution disabled unless explicitly enabled
- require the same authenticated mutation gate for outcome credit