How to generate structured output with the Ollama API

Structured output constrains an Ollama response to JSON or a JSON schema so application code can parse fields instead of scraping prose. It is useful for classification, extraction, routing, and any workflow where free-form text would create fragile downstream logic.

The format field can request generic JSON or provide a schema object. Pair the schema with a direct prompt and deterministic options when the result will be validated automatically.

The local fixture confirmed schema-shaped output with gpt-oss:20b. The same pattern works through api/chat and can be adapted to SDK calls that pass an equivalent schema.

1. Write the JSON schema that the response must match.

   {
     "type": "object",
     "properties": {
       "name": {"type": "string"},
       "capital": {"type": "string"}
     },
     "required": ["name", "capital"]
   }

2. Send the schema in the format field of a chat request.

   $ curl -s http://localhost:11434/api/chat -d '{
     "model": "gpt-oss:20b",
     "messages": [{"role":"user","content":"Return a JSON object for Canada with keys name and capital."}],
     "stream": false,
     "format": {
       "type":"object",
       "properties":{"name":{"type":"string"},"capital":{"type":"string"}},
       "required":["name","capital"]
     },
     "options": {"temperature": 0}
   }'
   {"message":{"content":"{\"name\":\"Canada\",\"capital\":\"Ottawa\"}"},"done":true}

3. Parse and validate the returned content before using it.

   $ python3 - <<'PY'
   import json
   content = '{"name":"Canada","capital":"Ottawa"}'
   obj = json.loads(content)
   assert set(["name", "capital"]).issubset(obj)
   print(obj["capital"])
   PY
   Ottawa

4. Keep reasoning fields separate from the structured content when using thinking models.

   $ jq '.message.content' response.json
   "{\"name\":\"Canada\",\"capital\":\"Ottawa\"}"

   Do not parse message.thinking as application data.

5. Reject malformed or missing fields in the caller.

   $ python3 - <<'PY'
   import json
   required = {"name", "capital"}
   obj = json.loads('{"name":"Canada"}')
   print(required <= obj.keys())
   PY
   False