Add support for specifying model (Ollama OpenAI-compatible API requires it) #2

New issue

Open

opened 2026-01-24 16:12:57 -06:00 by flome00 · 0 comments

flome00 commented

2026-01-24 16:12:57 -06:00

Hi! Thanks for Sortana — it’s a great add-on.

I’m trying to use Sortana with Ollama via its OpenAI-compatible API. The endpoint is reachable and works fine when model is provided, but Sortana appears not to send a model field in the request (or there is no way to configure it). Ollama then returns an error and Sortana does not classify/match.

Environment

Thunderbird: 140.6.0esr (64-Bit)
Sortana: 2.2.0
Ollama: 0.13.5
Model: qwen2.5:14b
Ollama host: http://ollama.lan:11434

Expected behavior

Sortana should be able to call OpenAI-compatible endpoints like Ollama by including model in the payload (or providing a setting for it).

Actual behavior

Ollama responds with an error because the request has no model (or model is empty), e.g.:

{"error":{"message":"model '' not found","type":"api_error","param":null,"code":null}}

Sortana then does not classify / no actions are triggered.

Reproduction

Configure Sortana endpoint to http://ollama.lan:11434
Create a rule (e.g. “is this email a newsletter?”) and try to run it on selected messages
Ollama returns an error because model is missing.

Proof that Ollama works with `model`

Calling Ollama directly works as expected if model is set:

curl -s http://ollama.lan:11434/v1/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen2.5:14b",
    "prompt": "Return ONLY valid JSON: {\"match\": true, \"reason\": \"ok\"}",
    "temperature": 0,
    "max_tokens": 50
  }'

This returns a normal completion response.

Suggested fix

Add a Settings field for the model name and include it in the request payload when calling /v1/completions (and/or /v1/chat/completions if supported).

Example (pseudo-code):

// read from settings (new option)
const model = settings.model?.trim();

// build request payload
const payload = {
  // existing fields
  prompt,
  temperature,
  max_tokens,
  // ...
};

// Ollama/OpenAI-compatible servers need this
if (model) {
  payload.model = model;
}

Optionally, default to something sensible if the user selects a “local OpenAI-compatible endpoint”:

payload.model ??= "qwen2.5:14b";

Bonus improvement (optional)

Some OpenAI-compatible servers primarily expect chat format. If Sortana ever supports chat completions, it would also be helpful to allow selecting between:

/v1/completions (prompt)
/v1/chat/completions (messages)

But the minimal change needed for Ollama is simply to send model.

Thanks again — happy to test a build/PR if needed.

Hi! Thanks for Sortana — it’s a great add-on. I’m trying to use Sortana with **Ollama** via its **OpenAI-compatible API**. The endpoint is reachable and works fine when `model` is provided, but Sortana appears not to send a `model` field in the request (or there is no way to configure it). Ollama then returns an error and Sortana does not classify/match. ### Environment * Thunderbird: 140.6.0esr (64-Bit) * Sortana: 2.2.0 * Ollama: 0.13.5 * Model: `qwen2.5:14b` * Ollama host: `http://ollama.lan:11434` ### Expected behavior Sortana should be able to call OpenAI-compatible endpoints like Ollama by including `model` in the payload (or providing a setting for it). ### Actual behavior Ollama responds with an error because the request has no model (or model is empty), e.g.: ```json {"error":{"message":"model '' not found","type":"api_error","param":null,"code":null}} ``` Sortana then does not classify / no actions are triggered. ### Reproduction 1. Configure Sortana endpoint to `http://ollama.lan:11434` 2. Create a rule (e.g. “is this email a newsletter?”) and try to run it on selected messages 3. Ollama returns an error because `model` is missing. ### Proof that Ollama works with `model` Calling Ollama directly works as expected if `model` is set: ```bash curl -s http://ollama.lan:11434/v1/completions \ -H "Content-Type: application/json" \ -d '{ "model": "qwen2.5:14b", "prompt": "Return ONLY valid JSON: {\"match\": true, \"reason\": \"ok\"}", "temperature": 0, "max_tokens": 50 }' ``` This returns a normal completion response. ### Suggested fix Add a **Settings** field for the model name and include it in the request payload when calling `/v1/completions` (and/or `/v1/chat/completions` if supported). Example (pseudo-code): ```js // read from settings (new option) const model = settings.model?.trim(); // build request payload const payload = { // existing fields prompt, temperature, max_tokens, // ... }; // Ollama/OpenAI-compatible servers need this if (model) { payload.model = model; } ``` Optionally, default to something sensible if the user selects a “local OpenAI-compatible endpoint”: ```js payload.model ??= "qwen2.5:14b"; ``` ### Bonus improvement (optional) Some OpenAI-compatible servers primarily expect chat format. If Sortana ever supports chat completions, it would also be helpful to allow selecting between: * `/v1/completions` (prompt) * `/v1/chat/completions` (messages) But the minimal change needed for Ollama is simply to send `model`. Thanks again — happy to test a build/PR if needed.