~/tools / Run a local LLM on your server / Ollama OpenAI Converter

Ollama OpenAI Converter — Convert chat requests and responses between Ollama /api/chat and /v1/chat/completions

Convert a chat request between Ollama's native /api/chat and its OpenAI-compatible /v1/chat/completions, in either direction, and turn an /api/chat response into the response that /v1/chat/completions returns. The conversion follows Ollama's own source code, and the tool lists every field that is renamed, dropped or has no counterpart. Nothing leaves your browser.

curl http://localhost:11434/api/chat -d '{"model":"qwen2.5:7b","messages":[{"role":"system","content":"Answer in one sentence."},{"role":"user","content":"What is the weather in Toronto?"},{"role":"assistant","content":"","tool_calls":[{"id":"call_1","function":{"name":"get_weather","arguments":{"city":"Toronto"}}}]},{"role":"tool","content":"11 degrees, cloudy","tool_name":"get_weather","tool_call_id":"call_1"}],"stream":false,"options":{"stop":["END"],"num_predict":200,"temperature":0.2},"tools":[{"type":"function","function":{"name":"get_weather","description":"Weather for a city","parameters":{"type":"object","properties":{"city":{"type":"string"}},"required":["city"]}}}]}'

{
  "model": "qwen2.5:7b",
  "messages": [
    {
      "role": "system",
      "content": "Answer in one sentence."
    },
    {
      "role": "user",
      "content": "What is the weather in Toronto?"
    },
    {
      "role": "assistant",
      "content": "",
      "tool_calls": [
        {
          "id": "call_1",
          "function": {
            "name": "get_weather",
            "arguments": {
              "city": "Toronto"
            }
          }
        }
      ]
    },
    {
      "role": "tool",
      "content": "11 degrees, cloudy",
      "tool_name": "get_weather",
      "tool_call_id": "call_1"
    }
  ],
  "stream": false,
  "options": {
    "stop": [
      "END"
    ],
    "num_predict": 200,
    "temperature": 0.2
  },
  "tools": [
    {
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Weather for a city",
        "parameters": {
          "type": "object",
          "properties": {
            "city": {
              "type": "string"
            }
          },
          "required": [
            "city"
          ]
        }
      }
    }
  ]
}

Ollama answers chat requests on two endpoints: its native /api/chat and the OpenAI-compatible /v1/chat/completions. The second one does not run requests on its own. In Ollama's source code (openai/openai.go), each request is first converted into an /api/chat request, and the answer is converted back. This tool does the same conversion in your browser. Paste a request body or a response, and it detects which one it is. From the OpenAI form to Ollama: max_tokens becomes options.num_predict, a single stop string becomes a list, temperature, top_p, seed and the penalties move into options, response_format becomes format ("json", or the schema from json_schema.schema), reasoning_effort becomes think (none becomes false, minimal becomes low, xhigh and ultra become max), the arguments of tool calls change from a JSON string to an object, and a tool message gets the function name looked up from its tool_call_id. Each part of a content array becomes its own message, and images must be base64 data URLs of JPEG, PNG or WebP images; image URLs are rejected. Fields that are not in Ollama's code, such as tool_choice, n, user, logit_bias and max_completion_tokens, are dropped without an error. In our test on Ollama 0.40.11, a request with max_completion_tokens 3 still returned a long answer, because Ollama reads only max_tokens. In the other direction, options without an OpenAI counterpart, such as num_ctx, top_k and min_p, are listed instead of written. For a response, done_reason becomes finish_reason (stop becomes tool_calls when the model called a tool), prompt_eval_count and eval_count become usage, thinking becomes reasoning, and the durations are dropped. Ollama writes the id as chatcmpl- followed by a random number; the tool writes chatcmpl-0. We checked the conversion by sending both forms of each request to Ollama 0.40.11 with the debugging field _debug_render_only, which returns the prompt the model would see, and comparing the two prompts, and by comparing converted responses with the real answers of /v1/chat/completions.

The core logic of this tool is open source under the MIT License. github.com/seigo5555/toolkitbox-lib/src/ollama-openai.ts