Skip to main content
Generate the next message in a chat with a provided model. This endpoint supports streaming responses and maintains conversation history through messages.

​
Request

​
Endpoint

POST /api/chat

​
Request Body

​
model
string
required
The model name to use for chat completion (e.g., llama3.2, mistral)
​
messages
array
required
Array of message objects representing the conversation history
​
stream
boolean
default:"true"
Enable streaming of response chunks. Set to false to wait for the complete response.
​
format
string | object
Format to return response in. Use "json" for JSON mode or provide a JSON schema object.
​
options
object
Model-specific options to customize inference behavior
​
keep_alive
string | number
default:"5m"
Duration to keep the model loaded in memory (e.g., "5m", "1h", or -1 for indefinite)
​
tools
array
List of tools the model can use for function calling
​
think
boolean | string
Enable thinking mode for reasoning models. Can be true/false or "high", "medium", "low"
​
truncate
boolean
default:"true"
Truncate chat history if prompt exceeds context length
​
shift
boolean
default:"true"
Shift chat history when hitting context length instead of erroring
​
logprobs
boolean
default:"false"
Return log probabilities for output tokens
​
top_logprobs
integer
default:"0"
Number of most likely tokens to return at each position (0-20). Requires logprobs: true

​
Response

​
Response Fields

​
model
string
The model name used for generation
​
created_at
string
Timestamp of when the response was created (ISO 8601 format)
​
message
object
The generated message
​
done
boolean
Whether the response is complete
​
done_reason
string
Reason for completion: stop, length, load, unload
​
total_duration
integer
Total time spent generating response (nanoseconds)
​
load_duration
integer
Time spent loading the model (nanoseconds)
​
prompt_eval_count
integer
Number of tokens in the prompt
​
prompt_eval_duration
integer
Time spent evaluating the prompt (nanoseconds)
​
eval_count
integer
Number of tokens generated
​
eval_duration
integer
Time spent generating response (nanoseconds)
​
logprobs
array
Log probability information for each token (when logprobs: true)

​
Examples

curl http://localhost:11434/api/chat -d '{
  "model": "llama3.2",
  "messages": [
    {
      "role": "user",
      "content": "Why is the sky blue?"
    }
  ],
  "stream": false
}'

​
Example Response

{
  "model": "llama3.2",
  "created_at": "2024-02-24T12:34:56.789Z",
  "message": {
    "role": "assistant",
    "content": "The sky appears blue due to Rayleigh scattering..."
  },
  "done": true,
  "done_reason": "stop",
  "total_duration": 5432109876,
  "load_duration": 123456789,
  "prompt_eval_count": 26,
  "prompt_eval_duration": 987654321,
  "eval_count": 298,
  "eval_duration": 4321098765
}

​
Streaming Example

curl http://localhost:11434/api/chat -d '{
  "model": "llama3.2",
  "messages": [
    {
      "role": "user",
      "content": "Tell me a story"
    }
  ]
}'

​
Error Responses

​
error
string
Error message describing what went wrong

​
Common Errors

  • 400 Bad Request: Invalid request body or model name
  • 404 Not Found: Model not found (need to pull it first)
  • 500 Internal Server Error: Server error during generation
When streaming is disabled ("stream": false), the complete response is returned as a single JSON object. With streaming enabled (default), responses are sent as newline-delimited JSON (NDJSON).

Build docs developers (and LLMs) love

Get started for freeTalk to us