Chat API

Pico AI Homelab supports the following OpenAI and Ollama-compatible endpoints for chat:

• v1/chat/completions OpenAI-compatible chat API

• api/chat Ollama-compatible chat API

• api/generate Ollama-compatible completion API

These endpoints are conform OpenAI and Ollama. Pico supports message content types text, image_url, and video_url.

Pico supports both LLM and VLM models. To discover which models have been downloaded and are available to clients, use the models API.

Request

POST /v1/chat

POST /api/chat

POST /api/generate

Name	Type	Description
model	String	Name of the embeddings model, e.g. all-MiniLM-L6-v2
messages	Array of messages	String or strings to embed
stream	Optional boolean	If true or nil, the response will be streamed to the client per token
reasoning	Optional enum	See reasoning
chat_template_kwargs	Optional dictionary	See reasoning
max_tokens	Optional integer	Deprecated, use max_completion_tokens instead.
max_completion_tokens	Optional integer	Upper bound for the number of tokens that can be generated for a completion, including visible output tokens and reasoning tokens
temperature	Optional float	What sampling temperature to use
frequency_penalty	Optional float
top_p	Optional float
user	Optional string	Ignored by Pico
format	Optional string	Ignored by Pico
options	Optional object	See Ollama options
think	Optional boolean	Enables or disables reasoning (conform Ollama API). Available from 1.1.18

Reasoning

Pico supports the OpenAI, vLLM, and Ollama way of enabling and disabling reasoning. The reasoning (OpenAI API-compatible) and enable_thinking options (vLLM API-compatible) are supported in Pico 1.1.14 and later, the think option (Ollama API-compatible) from 1.1.18.

Note that setting the reasoning options is optional. If no reasoning option is supplied, the model will use its default settings, which is often reasnoning enabled.

OpenAI-compatible Reasoning

Unlike OpenAI, Pico does not implement reasoning levels (low, medium, high); only binary ON or OFF states are supported. Requests specifying low will be interpreted by Pico as Reasoning = Off. Note that Pico also supports non-OpenAI standard values on, off, and none .

Name	Type	Description
effort	Enum	See below
summary	Optional string	This property is ignored

Name	Description
low	Pico interprets this condition as: Reasoning mode disabled
medium	Pico interprets this condition as: Reasoning mode enabled
high	Pico interprets this condition as: Reasoning mode enabled
on	Reasoning mode enabled
off	Reasoning mode disabled
none	Reasoning mode disabled

vLLM-compatible Reasoning

Alternatively, use the vLLM API by setting key enable_thinking to true or false in chat_template_kwargs

Ollama-compatible Reasoning

think is an optional boolean.

As of Pico AI Server 1.1.18, the thinking field in the response is not supported yet, meaning that thinking responses will be streamed as a regular response.