Autonomous research and knowledge archiving.
Discover. Preserve. Understand.
OrionSearch is an AI-powered platform designed to continuously collect and organize information about topics you follow, preserve the sources behind it, and turn the resulting archive into readable, citation-backed digital magazines.
Project status: In development · Hackathon MVP
Information about fast-moving topics is scattered across news outlets, websites, and reports. Repeatedly searching, comparing coverage, and remembering what changed takes time. OrionSearch aims to maintain a reusable, evolving knowledge archive instead of producing only a one-off answer.
The archive is the product. The magazine is one way to explore it.
These are product goals; not all capabilities are implemented yet.
- Topic archives: Organize research around topics, industries, technologies, or events.
- Automated research: Discover sources and revisit topics on a configurable schedule.
- Traceable evidence: Preserve source links and available publication and retrieval dates.
- Incremental updates: Add newly discovered material while retaining research history.
- Knowledge exploration: Search and browse saved sources, related coverage, and changes over time.
- AI-generated magazines: Create readable, structured editions with citations to underlying evidence.
- Configure a topic archive and its research preferences.
- Discover and retrieve relevant public sources.
- Analyze and organize evidence while preserving source links.
- Store results so future research can build on previous runs.
- Generate a magazine-style view from the accumulated archive.
| Component | Planned technology | Responsibility |
|---|---|---|
| Web client | Next.js, React, TypeScript | Archive management, search, and magazine reading |
| API | C#, ASP.NET Core (.NET 10) | HTTP API and application orchestration |
| Research workflow | LangGraph and a configurable LLM | Research planning, source analysis, and synthesis |
| Data store | PostgreSQL, Entity Framework Core, Npgsql | Archives, sources, findings, and research history |
| Background processing | .NET worker with a PostgreSQL queue; recurring scheduler to be decided | Leased research execution and future recurring jobs |
| Infrastructure | Docker, Docker Compose, GitHub Actions | Local environment, CI/CD, and deployment |
The implementation may evolve as the team builds the MVP. AI and search providers should remain configurable, and generated claims should retain links to their supporting sources.
The ASP.NET Core API includes archive management, metadata topic/date filtering, EF Core migrations and queued research execution; see the backend README for setup and integration prerequisites. Native ASP.NET Core Identity accounts are implemented behind explicit configuration; email/password registration, immediate login, refresh and password changes are described in Native Identity. No confirmation email or external identity provider is required. Public account activation, frontend integration and the AI team's agreed research contract remain pending. Default startup rejects protected requests and leaves research processing disabled. The Render hosting record documents the API/database deployment separately from product readiness.
Never commit secrets, API keys, local .env files, or user data.
A separate .NET executor and results API now provide background execution, archive research history, findings with authorized evidence references and source details. They share the existing PostgreSQL queue and version grants; no second job store is introduced. Hosting activation, live AI synthesis, frontend screens and product acceptance remain separate gates.
The current branch includes a prototype research worker with four public-source adapters (Google News RSS, arXiv RSS, the Hacker News Algolia API, and the Hugging Face blog feed), article filtering/deduplication, Gemini-assisted research planning and analysis, Markdown/JSON report output, and a FastAPI interface for synchronous research and background-job status.
The worker implements the research v1 contract consumed by the .NET executor. PostgreSQL queue ownership and ingestion remain in .NET; Python receives no database credentials. Discovery is the default. An explicit optional capture stage fetches public HTML/text and uploads immutable versions to private S3-compatible storage; .NET verifies actual bytes before ingestion and authorizes reads per archive. See capture storage for configuration and limits. Synthesis, LangGraph and real hosted end-to-end acceptance remain pending. See research integration.
From the repository root:
python -m venv .venv
# Activate the virtual environment for your shell, then:
python -m pip install -r services/research-worker/requirements-dev.txt
cd services/research-worker
python -m pytest tests agent/tests -qThe tests mock external services and should not require live source APIs or a Gemini key. Generated test output should use temporary directories rather than the repository's output folder.
From services/research-worker, set ORION_WORKER_API_KEY and GEMINI_API_KEY in your environment (or a local ignored .env file), then run:
python -m uvicorn api.main:app --host 127.0.0.1 --port 8000The integrated API exposes GET /health and POST /research. Research requires Authorization: Bearer <ORION_WORKER_API_KEY> and the versioned v1 JSON request; it does not accept the old topic/days body. The .NET API remains responsible for public research jobs and their persistent status. The old prototype routes are under /prototype/research and /prototype/research/jobs, require ORION_ENABLE_PROTOTYPE_API=true and X-API-Key, and stay disabled in integrated mode.
The integrated route writes no local reports. Standalone CLI/prototype JSON and Markdown reports remain under services/research-worker/agent/output/; these local artifacts are ignored by Git and must not be committed.
- Create a feature branch for each Jira issue.
- Keep pull requests focused and link the relevant
ORION-...issue. - Add or update tests for changed behavior.
- Once CI checks are configured, do not merge failing checks.
- Document new configuration fields and service dependencies.
The license has not yet been selected. No open-source license is implied until a LICENSE file is added.
OrionSearch — Knowledge that grows with time.