🦙 feat(providers): Ollama LLM provider — chat, embedding, streaming, function calling

[sbharatjoshi]

Summary

Implement the Ollama LLM provider — connecting Spector to locally-running Ollama models with chat completion, embedding generation, streaming, and function calling support. This is the default provider for local development.

---

🏛️ Technical Specifications

• SPI: Implements ProviderFactory via ServiceLoader (ADR-003)
• Module: spector-providers → com.spectrayan.spector.provider.ollama
• Key Classes:
  • OllamaProviderFactory — ServiceLoader entry point, creates OllamaProvider instances
  • OllamaProvider — implements LlmProvider interface: chat(), chatStream(), embed()
  • OllamaChatClient — HTTP client wrapping Ollama's REST API
• Ollama API Endpoints:
  • Chat: POST /api/chat with {"model": "...", "messages": [...], "stream": true}
  • Embedding: POST /api/embed with {"model": "...", "input": "..."}
  • Models: GET /api/tags (list available models)
• HTTP Client: java.net.http.HttpClient with virtual thread executor
• Streaming: Parse NDJSON (newline-delimited JSON) response for streaming chat
• Configuration: OLLAMA_BASE_URL (default: http://localhost:11434), SPECTOR_OLLAMA_MODEL (default: llama3.2)
• Function Calling: Map AgentTool definitions to Ollama's tool format (supported in Ollama 0.5+)

✅ Acceptance Criteria

• Given OLLAMA_BASE_URL is set, when the provider initializes, then it connects to Ollama and verifies model availability
• Given chat completion is requested, when Ollama is called, then responses are returned
• Given streaming is enabled, when tokens are generated, then they stream via SSE
• Given function calling is used, when tools are defined, then Ollama's native tool format is used
• Given embedding is requested, when embed() is called, then embeddings are returned as float[]

🚀 Contributor Guidance

1. Implement in providers/spector-providers module
2. Create META-INF/services/com.spectrayan.spector.provider.spi.ProviderFactory for ServiceLoader
3. Use java.net.http.HttpClient for all HTTP calls
4. Parse NDJSON streaming: read line-by-line, parse each line as JSON
5. Reference the ProviderFactory SPI from spector-provider-spi
6. Build: mvn clean compile -pl providers/spector-providers

📚 References

• Ollama API
• spector-provider-spi for the ProviderFactory interface

RICE-C Score

Factor	Score	Rationale
Reach	5/5	Default local provider
Impact	3x	Core functionality
Effort	M	~2 days

[github-actions]

This issue has been automatically marked as stale because it has not had activity in 30 days.

If you're still working on this or it's still relevant, please leave a comment and we'll keep it open. Otherwise, it will be closed in 14 days.

No pressure — life happens! You can always reopen or create a new issue later. 🙂

[jarvispectrayan]

Verified complete in codebase via commit 07fae767 (feat(synapse): add Provider SPI, registry, and management API). OllamaLlmProvider with chat, embedding, streaming, and function calling is fully implemented. Closing as DONE.