Opentrace is a RAG (Retrieval-Augmented Generation) platform that lets you upload documents and ask an AI assistant questions about them. Every answer is grounded in your documents with citations.
PDF, DOCX, HTML files, and any public web URL. See System Requirements for full details.
There is no hard file size limit, but very large files (hundreds of megabytes) may take longer to process. Processing time scales with the number of pages/content, not file size.
Yes! Create as many projects as you want. Each project has its own isolated knowledge base, conversations, and settings.
Responses are grounded in your documents — the AI can only use information that exists in your uploaded content. The citation system lets you verify every claim. Accuracy depends on document quality and the RAG strategy you choose.
Start with Basic (fastest). If results aren't satisfactory, try Hybrid for keyword + semantic search. Use Multi-Query for complex research questions. See RAG Strategies.
Yes, if you enable the Agentic agent type. The supervisor agent can route queries to both your documents and a web search engine (Tavily or DuckDuckGo). See Agent Types.
Documents are stored in AWS S3 with access controlled via presigned URLs. Document data in the database is scoped to your user account — no other user can access your documents or conversations.
No. Opentrace uses the OpenAI API, which does not use API-submitted data for model training as per OpenAI's data usage policy.
All data is permanently deleted: documents, S3 files, chunks, embeddings, conversations, and messages. This action cascades through the database and cannot be undone.
Yes — Opentrace is fully open-source. See the Self-Hosting section for Docker, local development, and AWS deployment guides.
At minimum: OpenAI API key, Supabase instance, AWS S3 bucket, Clerk account, and Redis. Tavily and ScrapingBee are optional. See Environment Variables.
