3 Commits (7cbb2853ced0b47e5957a4e7ec1607c03738fce0)

Author SHA1 Message Date
oabrivard 8a18b70aff fix: set max output tokens to 16384 for all LLM providers
OpenAI's default output limit (4096 tokens) was too low for structured
synthesis output with multiple categories and articles per category,
causing truncated JSON. Set 16384 for both OpenAI APIs (Responses +
Chat Completions) and Gemini. Anthropic already had 16384.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3 months ago
oabrivard aa6f1ba76b Phase 5: Generation pipeline with SSE progress, syntheses CRUD
Backend:
- Full 2-pass generation pipeline: LLM search -> URL scraping -> LLM rewrite
- Async generation with tokio::spawn, JobStore with per-user concurrency limit
- SSE progress streaming via axum::response::Sse + tokio::sync::watch
- Syntheses CRUD: list (paginated), get (ownership check), delete
- Prompt construction ported from original geminiService.ts
- Parallel URL scraping with bounded concurrency (max 10)
- Graceful partial failure handling (some URLs fail -> continue)
- 36 new unit tests, 16 integration tests

Frontend:
- Home dashboard: synthesis card grid, week badges, delete with confirmation
- Generate page: SSE-driven progress bar, step checklist, auto-redirect
- Synthesis detail: section-by-section display, external links, delete
- SSE client helper with auto-reconnect (exponential backoff)
- Date utilities with French locale formatting

Critical fixes applied:
- SSE EventSource now sends credentials (withCredentials: true)
- Gemini error logging sanitized to prevent API key leak in logs

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3 months ago
oabrivard 439e547367 Phase 4: LLM provider abstraction with Gemini, user API key encryption
Backend:
- LlmProvider async trait with generate_search_pass/generate_rewrite_pass
- GeminiProvider: googleSearch grounding (pass 1), structured JSON output (pass 2)
- AES-256-GCM encryption for user API keys at rest (per-key random nonces)
- MasterKey with zeroize-on-drop (no Clone to prevent unzeroized copies)
- User API key endpoints: list (prefix only), create/update, delete, test
- Dynamic category schema builder for structured LLM output
- Provider factory (Gemini implemented, OpenAI/Anthropic stubbed for Phase 6)
- 37 new unit tests (encryption, schema, Gemini serialization, factory)
- 17 integration tests (CRUD, encryption verification, ownership isolation)

Frontend:
- ApiKeyManager component: per-provider key management in Settings
- Password input with show/hide toggle, key prefix display (monospace)
- Test button validates key with minimal LLM call
- Status badges (configured/not configured)
- 11 new tests

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
3 months ago