Feature: Implement Ollama Provider Functionality

## Feature: Implement Ollama Provider Functionality

**Category**: Feature  
**Priority**: High  
**Related**: Providers module (crates/infrastructure/providers-ollama)

---

## What

Implement the `complete()` and `stream()` methods for the Ollama provider to enable actual API requests to Ollama instances.

---

## Why

The Ollama provider is currently configured in the gateway but returns "not yet implemented" errors. Users want to use local Ollama instances as providers through the Cortex gateway.

**Current state:**
```rust
async fn complete(&self, _req: &CompletionRequest) -> NuxaResult<CompletionResponse> {
    Err(NuxaError::provider("Ollama provider not yet implemented"))
}
```

**Expected behavior:**
- Forward requests to Ollama's `http://localhost:11434/api/generate` (or `/api/chat/generate` for chat models)
- Parse Ollama's response format
- Convert to Cortex's CompletionResponse format
- Handle streaming responses
- Respect timeout configuration

---

## What's Already There

- `OllamaProviderConfig` with `base_url`, `models`, `timeout_secs`
- Client initialization with timeout
- Basic health check (returns "not supported")
- Model validation in `supported_models()`

---

## What Needs to Be Done

### 1. Update Provider Implementation

In `crates/infrastructure/providers-ollama/src/lib.rs`:

#### A. Implement `complete()`

```rust
async fn complete(&self, req: &CompletionRequest) -> NuxaResult<CompletionResponse> {
    // Convert Cortex CompletionRequest -> Ollama API request
    // Call Ollama API endpoint
    // Parse response
    // Convert to CompletionResponse
    // Return result
}
```

**API endpoint:** `/api/generate` (or `/api/chat/generate` for chat models)

**Response format to handle:**
```json
{
  "model": "llama3.2",
  "created_at": "2024-01-15T...",
  "response": "Hello!",
  "done": true,
  "context": [...],
  "total_duration": 1234567890,
  "load_duration": 123456789,
  "prompt_eval_count": 10,
  "prompt_eval_duration": 123456789,
  "eval_count": 20,
  "eval_duration": 1111111101
}
```

#### B. Implement `stream()`

```rust
async fn stream(&self, req: &CompletionRequest) -> NuxaResult<futures::stream::BoxStream<'_, NuxaResult<StreamChunk>>> {
    // Similar to complete() but for streaming
    // Handle Ollama's chunked response
    // Convert to StreamChunk format
}
```

### 2. Request/Response Mapping

#### A. Input Mapping
- Map CompletionRequest fields to Ollama fields
- Handle system prompt, messages, temperature, max_tokens
- Map model name (handle aliases if needed)

#### B. Output Mapping
- Parse Ollama response into CompletionResponse
- Token usage (prompt_tokens, completion_tokens)
- Content and finish reason
- Timing information (optional)

### 3. Error Handling

- Handle connection timeouts (use configured timeout)
- Handle invalid model names
- Handle Ollama not running
- Return appropriate Cortex errors (ProviderError, TimeoutError)

### 4. Testing

- Mock Ollama server for tests
- Test with real Ollama if available
- Test streaming and non-streaming
- Test various model types (base, chat, code)

---

## Acceptance Criteria

- [ ] Ollama provider can complete chat completion requests
- [ ] Ollama provider can stream responses
- [ ] Timouts work correctly (respects `timeout_secs`)
- [ ] Invalid model names rejected gracefully
- [ ] Ollama not running detected and reported
- [ ] Parsing errors handled and reported
- [ ] Tests pass for all scenarios

---

## Notes

- Ollama API docs: https://github.com/ollama/ollama/blob/main/docs/api.md
- Consider using `/api/chat/generate` for chat models (if available)
- Ollama supports both `/api/generate` and `/api/chat/generate`
- Handle both JSON and chunked responses
- Consider caching for repeated requests (if upstream supports it)


Navigation Menu

Uh oh!

Feature: Implement Ollama Provider Functionality #65