Send HTTP POST notifications to configured webhook URLs when batch
jobs complete or when scheduled analysis detects significant changes.
- Add SPARC/webhooks.py with retry logic (3 attempts, exponential backoff)
- Support generic HTTP POST and Slack-compatible text payloads
- Integrate into batch job completion handler in api.py
- Configure via WEBHOOK_URLS env var (comma-separated)
- Payload includes event type, job ID, status, and summary
Closesleeworks-agents/SPARC#23
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add APScheduler-based background task that periodically re-analyzes
tracked companies and alerts on significant patent count changes.
- Add tracked_companies and alerts tables to database schema
- Add SPARC/scheduler.py with configurable interval and threshold
- Add admin endpoints: GET/POST/DELETE /admin/tracked, GET /admin/alerts
- Scheduler starts at app startup; interval via SCHEDULE_INTERVAL_HOURS
- Change threshold configurable via CHANGE_THRESHOLD_PERCENT env var
- apscheduler is optional; graceful fallback if not installed
Closesleeworks-agents/SPARC#22
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add GET /analytics/trends endpoint returning per-company analysis
counts by month and analysis type distribution over time. Render
these as a line chart (analyses per company) and stacked bar chart
(analysis types) on the Analytics page using recharts.
Closesleeworks-agents/SPARC#24
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add GET /export/{company_name} backend endpoint that returns analysis
records as a downloadable CSV file. Add Export CSV button to the
Analysis page that triggers the download via the API.
Closesleeworks-agents/SPARC#20
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add a cursor query parameter to GET /jobs and return a next_cursor
field in the response envelope. Existing clients using only limit
continue to work without modification. The cursor is an opaque token
encoding created_at and job_id for stable keyset pagination.
Closesleeworks-agents/SPARC#25
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Introduce a StorageBackend abstraction (local filesystem and S3) for
patent PDF storage. When STORAGE_BACKEND=s3, PDFs are read/written via
boto3 to an S3-compatible bucket instead of the local filesystem.
- Add SPARC/storage.py with LocalStorageBackend and S3StorageBackend
- Update serp_api.py save_patents and parse_patent_pdf to use storage
- Add storage config vars to config.py and .env.example
- Add optional MinIO service to docker-compose.yml (--profile s3)
- Add boto3 to requirements.txt
Closesleeworks-agents/SPARC#38
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When the PDF is not on disk, analyze_single_patent now looks up the
cached PDF link from the database and downloads it automatically.
If no link is cached, a clear FileNotFoundError is raised. Also adds
a GET /analyze/patent/{patent_id} API endpoint that exposes this
functionality and returns 404 when the PDF cannot be obtained.
Closesleeworks-agents/SPARC#36
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add module-level logger to serp_api.py with INFO-level messages for
patent queries and PDF downloads, and DEBUG-level messages for cache
hits and parsing details. All three target files (analyzer.py,
serp_api.py, llm.py) now use structured logging with no print() calls.
Closesleeworks-agents/SPARC#46
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add test job to build.yaml that runs pytest and ruff before building images
- Add standalone test.yaml workflow for PRs
- Add ruff.toml with E/F/I rules configured
- Fix all ruff lint errors: sort imports, remove unused imports, fix re-exports
- Build jobs now depend on test job passing (needs: test)
Closesleeworks-agents/SPARC#18Closesleeworks-agents/SPARC#19
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Replace get_db_client() creating new DatabaseClient on every call with a
module-level singleton initialized once at startup via init_db_client()
- Add init_db_client() and close_db_client() lifecycle functions called
from FastAPI lifespan handler
- Migrate all DatabaseClient methods from legacy self.connect()/self.conn
to pooled self.get_conn() context manager for thread-safe connection reuse
- Pool is properly torn down on application shutdown
Closesleeworks-agents/SPARC#7
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Make LLM model configurable via MODEL env var, default anthropic/claude-3.5-sonnet (#12)
- Expose SERP cache TTL as SERP_CACHE_TTL_HOURS env var, default 24 hours (#13)
- Fix Patent.patent_id type annotation from int to str in types.py (#14)
- Replace all print() calls with structured logging in analyzer.py and llm.py (#11)
- Add LOG_LEVEL config with basicConfig setup in config.py
- Add model and serp_cache_ttl_hours to config.py
Closesleeworks-agents/SPARC#11Closesleeworks-agents/SPARC#12Closesleeworks-agents/SPARC#13Closesleeworks-agents/SPARC#14
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add jobs table to database schema (job_id, status, progress, result_json, etc.)
- Add DatabaseClient methods: create_job, update_job, get_job, list_jobs
- Add mark_stale_jobs_failed() called at startup to handle interrupted jobs
- Refactor _run_batch_job and job endpoints to read/write from PostgreSQL
- Remove in-memory _jobs dict; job state now survives API restarts
- Update init_database.py to list all tables in output
Closesleeworks-agents/SPARC#8
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add docstring to analyze_single_patent explaining the PDF prerequisite
- Raise FileNotFoundError with helpful message when PDF is missing
- Add patent PDF storage section to README with Docker volume mount example
- Commit frontend/package-lock.json for reproducible builds
Closesleeworks-agents/SPARC#15Closesleeworks-agents/SPARC#17
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add check_jwt_secret() that refuses default JWT secret when APP_ENV != development
- Make CORS origins configurable via CORS_ORIGINS env var (comma-separated)
- Replace hardcoded postgres credentials in docker-compose.yml with env var references
- Add APP_ENV and cors_origins to config.py
- Update .env.example with all required variables and documentation
- Add tests for JWT startup guard and CORS configuration
Closesleeworks-agents/SPARC#4Closesleeworks-agents/SPARC#5Closesleeworks-agents/SPARC#6
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
_analyze_company_safe was calling SERP.query directly, bypassing the
SERP query cache in analyze_company. Now delegates fully to
analyze_company() and reads patent_count from the serp_queries cache.
Before querying SERP API, check serp_queries cache (24h TTL). Before
downloading/parsing each patent, check patents table for cached
minimized_content. Store results after processing so repeated analyses
skip all network I/O and PDF parsing entirely.
Replace the sequential per-patent loop with a ThreadPoolExecutor
(workers controlled by PATENT_THREAD_WORKERS config). Each patent is
processed independently in _process_single_patent, which is thread-safe
since SERP methods are stateless and operate on separate files.
The SERP query had a frozen date range (Oct-Nov 2025) that returned
stale patents. Now computes a rolling window from config
(PATENT_SEARCH_DAYS, default 90 days). Also adds filesystem-level PDF
caching to skip re-downloading existing patent PDFs, and adds
PATENT_THREAD_WORKERS config for upcoming parallel processing.
_analyze_company_safe called SERP.query() then passed the company name
to analyze_company() which called SERP.query() again — doubling API
usage. Now analyze_company() accepts an optional patents param so callers
can pass pre-fetched results through.
Add ROOT_PATH environment variable support so FastAPI generates correct
URLs for Swagger UI when served behind nginx at /api. This fixes the
"invalid version field" error when accessing /api/docs.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Change TokenPayload.sub type from int to str per JWT RFC 7519
- Add user_id property to TokenPayload for int conversion
- Update token creation to serialize user_id as string
- Update token consumers to use payload.user_id
- Change dashboard port from 3000 to 8080
- Add pydantic[email] for email validation
- Update default admin email to admin@sparc.dev🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Protect all analysis endpoints with JWT authentication:
- Require valid access token for analysis operations
- Add CORS middleware for React frontend (localhost:3000, 5173)
Add auth endpoints:
- POST /auth/register - user registration (first user becomes admin)
- POST /auth/login - JWT token issuance
- POST /auth/refresh - token refresh
- GET /auth/me - current user info
Add admin endpoints:
- GET /admin/users - list all users
- PATCH /admin/users/{id}/role - update user role
- DELETE /admin/users/{id} - delete user
Add analytics endpoint:
- GET /analytics - usage statistics by company and type
Update .env.example with USE_CACHE and JWT_SECRET config
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Replace USE_DATABASE toggle with USE_CACHE for smarter LLM response handling:
- Add prompt hashing for efficient cache lookups
- Cache API responses in database to reduce token usage
- Always store responses for analytics (cache or fresh)
Add user authentication infrastructure:
- User table with bcrypt password hashing
- CRUD operations for user management
- Role-based access control (admin/user)
Dependencies: add bcrypt and PyJWT for auth
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Not all Google Patents results include PDF download links. Previously
this caused a KeyError when accessing patent["pdf"]. Now patents
without PDF links are gracefully skipped with documentation explaining
when this occurs (recent filings, international patents, restricted access).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Create REST API with endpoints for single and batch analysis
- Add async job support for long-running batch operations
- Implement job status tracking and listing endpoints
- Add 9 tests for API endpoints
- Update requirements.txt with fastapi, uvicorn, httpx
- Document API usage in README
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add CompanyAnalysisResult and BatchAnalysisResult dataclasses
- Implement analyze_companies() for concurrent batch analysis
- Implement analyze_companies_sequential() for rate-limited scenarios
- Add progress callback support for monitoring batch jobs
- Include 5 new tests for batch processing functionality
- Fix pre-existing test mock issue in test_llm.py
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Implements a database mode that stores LLM prompts and responses in PostgreSQL
instead of making API calls. This enables:
- Testing without consuming API credits
- Collecting analytics on usage patterns
- Development and debugging workflows
Changes:
- Added DatabaseClient class for PostgreSQL operations
- Modified LLMAnalyzer to support database/API mode toggle
- Added USE_DATABASE config flag to switch between modes
- Included Docker Compose setup for PostgreSQL
- Added utility scripts for database init and analytics viewing
- Comprehensive documentation in DATABASE_MODE.md
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Replace direct Anthropic API integration with OpenRouter to enable
more flexible LLM provider access while maintaining Claude 3.5 Sonnet.
Changes:
- Replace anthropic package with openai in requirements.txt
- Update config to use OPENROUTER_API_KEY instead of ANTHROPIC_API_KEY
- Migrate LLMAnalyzer from Anthropic client to OpenAI client with
OpenRouter base URL (https://openrouter.ai/api/v1)
- Update model identifier to OpenRouter format: anthropic/claude-3.5-sonnet
- Convert API calls from messages.create() to chat.completions.create()
- Update response parsing to match OpenAI format
- Rename API key parameter in CompanyAnalyzer from anthropic_api_key
to openrouter_api_key
- Update all tests to mock OpenAI client instead of Anthropic
- Fix client initialization to accept direct API key parameter
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Created CompanyAnalyzer class that orchestrates the complete pipeline:
1. Retrieves patents via SERP API
2. Downloads and parses PDFs
3. Minimizes content (removes bloat)
4. Analyzes portfolio with LLM
5. Returns performance estimation
Features:
- Full company portfolio analysis
- Single patent analysis support
- Robust error handling (continues on partial failures)
- Progress logging for user visibility
Updated main.py with clean example usage demonstrating the high-level API.
Added comprehensive test suite (7 tests) covering:
- Full pipeline integration
- Error handling at each stage
- Single patent analysis
- Edge cases (no patents, all failures)
All 26 tests passing.
This completes the core functionality for patent-based company
performance estimation.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Implemented LLMAnalyzer class using Anthropic's Claude API for:
- Single patent content analysis
- Portfolio-wide analysis across multiple patents
- Configurable API key management via environment variables
Key features:
- Uses Claude 3.5 Sonnet for high-quality analysis
- Structured prompts for innovation assessment
- Token limits optimized per use case (1024 for single, 2048 for portfolio)
- Analyzes: innovation quality, market potential, strategic direction
Updated config.py to support ANTHROPIC_API_KEY environment variable.
Added comprehensive test suite (6 tests) covering:
- Initialization from config and direct API key
- Single patent analysis
- Portfolio analysis
- Token limit validation
All 19 tests passing.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Removed SPARC/patent_api.py as it contained duplicate implementations
of parse_patent_pdf, extract_section, and clean_patent_text functions
that are already present in SPARC/serp_api.py as static methods.
The serp_api.py implementation is actively used in main.py, while
patent_api.py was unused legacy code.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
agent-company
AI-Manager
0xWheatyz