Alan-Jowett / CoPilot-For-Consensus / 20285183912

78%

push

github

Replace scaffold summary text with real Ollama output (#307)

* Initial plan

* Implement real Ollama API integration in LocalLLMSummarizer

- Replace placeholder text with actual Ollama API calls via requests library
- Add requests dependency to setup.py
- Implement proper error handling for timeouts and connection errors
- Update tests to mock API calls and verify real content flow
- Fix config schema default for llm_backend (ollama -> local for consistency)

Co-authored-by: Alan-Jowett <20480683+Alan-Jowett@users.noreply.github.com>

* Add integration test for real Ollama content flow

- Verify real LLM-generated content (not placeholder) flows through the pipeline
- Test LocalLLMSummarizer integration with SummarizationService
- Assert placeholder text is absent from generated summaries
- Confirm SummaryComplete events contain actual LLM output

Co-authored-by: Alan-Jowett <20480683+Alan-Jowett@users.noreply.github.com>

* Address code review feedback

- Move requests import to module level in tests
- Add explicit Content-Type header to Ollama API call for better compatibility
- All tests still passing after fixes

Co-authored-by: Alan-Jowett <20480683+Alan-Jowett@users.noreply.github.com>

* feat: Add persistent Ollama model caching with local LLM default

- Change Ollama storage from Docker volume to bind mount (./ollama_models/)
  to persist models across 'docker compose down -v'
- Add ollama-model-loader service to auto-pull mistral model on startup
- Set LLM_BACKEND default to 'local' for development (real LLM)
- Configure CI to use 'mock' backend for fast, reliable tests
- Add ollama_models/ to .gitignore
- Update orchestrator and summarization to depend on model loader
- Remove test artifacts (test-results.xml)
- Skip complex integration test with clear justification

This enables real LLM-based summarization in local development while
keeping CI fast. First run downloads 4.4GB mistral model; subsequent
runs are instant due to local caching... (continued)

25 of 28 new or added lines in 1 file covered. (89.29%)

4098 of 4888 relevant lines covered (83.84%)

0.95 hits per line