Chat Completion
mistral.chat_completion
ActionAI & modelsconnection required
Calls the Mistral AI API with the connection's credentials.
Send a prompt or a full message list to a Mistral chat model and return the assistant's answer together with the token usage.
Connection
- Required
- Yes — the node will not save without one
- Auth type
- API key
- Providers
- Mistral AI
- Secret shape
- {"api_key": "…"} — a Mistral API key
- Config it needs
- base_url, default_model
- Resolved by
- config.connection_id, else node.connection_id
Configuration
13 fields, 1 required. A dotted path is a nested object and name[] an array of objects — the editor renders the same tree.
| Field | Type | Required | Default | Constraints | Example |
|---|---|---|---|---|---|
model Chat model to call, e.g. mistral-large-latest or mistral-small-latest. | string | Yes | — | — | mistral-large-latest |
prompt The user message, when you are not supplying a full message list. | string | No | — | — | Summarise this support ticket in two sentences. |
system Optional system instruction prepended to the prompt. | string | No | — | — | — |
messages A full conversation as [{role, content}] objects; overrides prompt/system. | array | No | — | array of object | — |
temperature Sampling temperature. Lower is more deterministic; Mistral recommends staying at or below 0.7 for most tasks. | number | No | — | 0–1.5 | — |
max_tokens Upper bound on the tokens generated in the answer. | number | No | — | ≥ 1 | — |
top_p Nucleus-sampling mass to consider. Use this or temperature, not both. | number | No | — | 0–1 | — |
json_mode Ask the model to reply with a single JSON object (response_format json_object). | boolean | No | false | — | — |
tools Tool definitions in Mistral's function-calling format, as JSON. | array | No | — | array of object | — |
tool_choice How the model may use the supplied tools. | string | No | — | one of "auto", "any", "none", "required" | — |
stop Stop sequences: generation ends when one of these appears. | array | No | — | array of string | — |
random_seed Seed for sampling, so repeated calls with identical input agree. | number | No | — | ≥ 0 | — |
safe_prompt Prepend Mistral's built-in safety system prompt to the conversation. | boolean | No | false | — | — |
- At least one of prompt, messages must be set.
Example configuration
Every value below is the contract’s own example or default — nothing invented.
{
"json_mode": false,
"model": "mistral-large-latest",
"prompt": "Summarise this support ticket in two sentences.",
"safe_prompt": false
}Output
One shape.
result
| Field | Type | Example |
|---|---|---|
id Mistral's id for this completion. | string | — |
model The model that actually answered. | string | — |
content The assistant message text. | string | — |
role Always 'assistant' for a completion. | string | — |
finish_reason Why generation stopped, e.g. stop or length. | string | — |
tool_calls Tool calls the model asked for, when tools were supplied. | array | — |
prompt_tokens Tokens consumed by the prompt. | number | — |
completion_tokens Tokens generated in the answer. | number | — |
total_tokens Prompt plus completion tokens. | number | — |
choices The raw choices array, for the n>1 or logprobs cases. | array | — |
{
"choices": [],
"completion_tokens": 0,
"content": "",
"finish_reason": "",
"id": "",
"model": "",
"prompt_tokens": 0,
"role": "",
"tool_calls": [],
"total_tokens": 0
}See Chat Completion run in a workflow.
In a technical demo we add the Mistral AI connection, drop the node on the canvas and let Workflow Health pre-flight it before the first real run.