Inline CLI — Command Reference
The Inline CLI lets you embed :command tokens directly inside your chat prompt to dynamically control API settings — model selection, internet search, memory, response shaping, and integrity — all without any extra API calls.
How to enable: include inline_cli: "all" (or a comma-separated list of feature groups) in your /v1/chat/completions request body. Then embed :command tokens anywhere in your prompt text. Commands are parsed before the LLM call and applied to the request.
Feature groups: model, shaping, integrity, internet, memory, information
State integration: Commands that include set (e.g. :setmodel, :setmemoryon, :setsearchlang) automatically persist their changes to the API key state via the State API, so subsequent requests inherit the updated configuration without re-sending the command.
NLP mode: When NLP is enabled in state, the LLM itself can interpret natural language instructions and emit the appropriate :command tokens, enabling fully conversational control of API settings by your agent.
Information Commands
Direct commands — the prompt is not forwarded to the LLM
| Command | Description |
|---|---|
| :getstate | Returns the current API key state as a JSON response. |
| :help | Returns this command reference. |
Model Commands
Switch which model handles this request, persistently or for one query
| Command | Description |
|---|---|
| :setmodel:<provider>/<model> | Persistently set the model (saved to state). |
| :unsetmodel | Clear the persistent model override. |
| :answerwith<name> | Use a named model for this query only. Options: openai, gpt, anthropic, claude, grok, llama, gemini, deepseek, mistral, mixtral, smart, cheap, coding. |
Shaping Commands
Adjust temperature, top_p, and penalty settings
| Command | Description |
|---|---|
| :beprecise | temperature=0.2, top_p=0.5 |
| :bebalanced | temperature=0.5, top_p=0.8 |
| :becreative | temperature=0.9, top_p=0.95 |
| :becrazy | temperature=1.5, top_p=1.0 |
| :becoder | temperature=0.1, top_p=0.4, penalties=0 |
| :avoidrepeat | frequency_penalty=1.2 |
| :answerdiverse | presence_penalty=1.2 |
| :stayontopic | presence_penalty=0, frequency_penalty=0 |
Integrity Commands
Reduce hallucinations by querying multiple times and selecting the best answer
| Command | Description |
|---|---|
| :unsetintegrity | Disable integrity mode. |
| :setintegrity | Enable integrity (best of 2) — saved to state. |
| :setsuperintegrity | Enable super integrity (best of 3) — saved to state. |
| :answerintegrity | Enable integrity for this query only. |
| :answersuperintegrity | Enable super integrity for this query only. |
Internet Search Commands
Augment the prompt with live web search results before sending to the LLM
| Command | Description |
|---|---|
| :search | Low-depth web search (1 query, 1 result). |
| :searchmore | Medium-depth web search (2 queries, 2 results each). |
| :deepsearch | High-depth web search (4 queries, 2 results each). |
| :setsearchlang:<code> | Set search language (2-letter ISO code, e.g. en). Saved to state. |
| :setsearchgeo:<code> | Set search geography (2-letter country code, e.g. US). Saved to state. |
Memory Commands
Control conversation memory (stored in vector DB, keyed by session)
| Command | Description |
|---|---|
| :setmemoryon | Enable memory for this session. Saved to state. |
| :setmemoryoff | Disable memory. Saved to state. |
| :clearmemory | Clear all stored memory for the current session. |
| :setmemexpire:<minutes> | Set memory TTL in minutes (5–1440). Saved to state. |
Request
- Base URL: https://apipie.ai
- URL: https://apipie.ai/v1/chat/completions/inline-cli
- Auth: one of:
- HTTP bearer
- API key in header x-api-key
Response
This path is documentation-only. Use POST /v1/chat/completions with inline_cli enabled.
Changes
No recorded changes to this operation across all 1 revision of this API.