[Feature request] Ollama API compatibility /api/tags

[grigio]

Describe the bug
Some software do not provide an openai-compatible options but they implement Ollama, is possible to have a /api/tags implementation?

Expected behaviour

# ollama endpoint
curl http://ollama.lan:11434/api/tags
{"models":[{"name":"gemma3n:e4b","model":"gemma3n:e4b","modified_at":"2025-06-27T13:46:49.045595241Z","size":7547589116,"digest":"15cb39fd9394fd2549f6df9081cfc84dd134ecf2c9c5be911e5629920489ac32","details":{"parent_model":"","format":"gguf","family":"gemma3n","families":["gemma3n"],"parameter_size":"6.9B","quantization_level":"Q4_K_M"}},{"name":"qwen3:30b-a3b","model":"qwen3:30b-a3b","modified_at":"2025-04-29T10:27:38.117801176Z","size":18622562938,"digest":"2ee832bc15b549ed17a6b83de0e0d3eab226b5e727ca046d22e11a2b50d9cd0c","details":{"parent_model":""..

# llama-swap endpoint
curl http://llama-swap.lan:9292/api/tags 
<empty response>

[mostlygeek]

check out llama-swappo, which adds an ollama translation layer.

[grigio]

Thanks for the reply, will it merged in llama-swap in the future?

[mostlygeek]

The plan is to keep them separate for now.

[mostlygeek]

Here is the PR (was not merged) that eventually led to llama-swappo: #147. My rational hasn't changed.

[paccerdk]

I sincerely hope you'll (re)consider this @mostlygeek.
This is a native feature of llama-server: ggml-org/llama.cpp#13659

A feature which is now effectively "filtered out" when running llama-swap vs llama-server natively
- this seems to go against the original goal of this project

As an example, that means users can't easily run Github Copilot with llama-swap, as you're probably already aware

[mostlygeek]

@paccerdk,

AFAICT, The /api/tags and /api/chat are not compatible with /v1/models and /v1/chat/completions request and response formats. I haven't looked at llama.cpp's implementation but simply pointing /tags/chat at the v1/chat/completions handler would break clients.

I wouldn't mind being wrong here because adding the endpoints would be trivial.

[paccerdk]

Sorry, I'm not sure i understand your question - maybe you misunderstood what i meant by "native feature"?
The PR I'm linking to discusses the implementation of the /api/tags and /api/chat endpoints (natively supported) in llama-server.

I didn't mean to imply that its simply a matter of pointing /tags/chat to v1/chat/completions if that's what you understood from my post

[mostlygeek]

@paccerdk I took a look at ollama's api documentation and the /api/chat endpoint does not appear to send back responses that are the same as the openAI compatible /v1/chat/completions.

I think llama.cpp's implementation is incorrect; but I haven't looked into it too deeply. If someone could confirm llama-server is a drop in replacement for /api/chat I'd be more willing to look into support that endpoint.

[paccerdk]

You might be right about /api/chat - it seems give a 404 with the current llama.cpp build (i don't know if it worked on a previous build).

I did some quick testing with copilot + native llama-server, and copilot does work - It seems that having /api/tags is enough for at least basic functionality

In case it helps, i see the following requests in llama-server logs:

srv  log_server_r: request: GET /api/events 127.0.0.1 404
srv  log_server_r: request: GET /api/tags 127.0.0.1 200
srv  log_server_r: request: GET /api/chat 127.0.0.1 404
srv  log_server_r: request: POST /v1/chat/completions 127.0.0.1 200

And this is what i get from /api/tags:

{
  "models": [
    {
      "name": "Qwen3-Coder",
      "model": "Qwen3-Coder",
      "modified_at": "",
      "size": "",
      "digest": "",
      "type": "model",
      "description": "",
      "tags": [
        ""
      ],
      "capabilities": [
        "completion"
      ],
      "parameters": "",
      "details": {
        "parent_model": "",
        "format": "gguf",
        "family": "",
        "families": [
          ""
        ],
        "parameter_size": "",
        "quantization_level": ""
      }
    }
  ],
  "object": "list",
  "data": [
    {
      "id": "Qwen3-Coder",
      "object": "model",
      "created": 1754417888,
      "owned_by": "llamacpp",
      "meta": {
        "vocab_type": 2,
        "n_vocab": 151936,
        "n_ctx_train": 262144,
        "n_embd": 2048,
        "n_params": 30532122624,
        "size": 17659361280
      }
    }
  ]
}

I'm unsure about the impact of missing /api/events and /api/chat endpoints on functionality.

I wish Copilot didn't use Ollama-specific implementation, but that's a different topic.

You can choose to support this with /api/tags (like llama-server), a more complete Ollama API, or not at all - or perhaps even consider allowing custom endpoints instead somehow.

I primarily wanted to clarify that llama-server does support some of these endpoint (s), so you can make an informed decision and not just reject it because it seemed Ollama-specific.

I honestly don't know how many popular apps rely on Ollama's API, and hopefully VS Code will eventually support standard OpenAI API (there's a relevant issue with 200+ upvotes: microsoft/vscode-copilot-release#7518).

Either way, thanks for excellent software!