Seahorse
Your documents, alive.
What is Seahorse?
Seahorse is a universal document engine that turns any collection of documents into a living, self-improving knowledge system. Connect your existing vector database, upload PDFs directly, or edit individual documents with cognitive analysis as inline annotations.
Most teams accumulate documents and forget about them. Over time, content becomes stale, contradictions creep in, orphan documents sit isolated with no connections, and coverage gaps grow silently. Nobody notices until quality degrades.
Seahorse watches your documents continuously. It detects problems, validates observations through adversarial auditing, learns patterns, applies fixes, and gets smarter with every cycle.
Three Ways to Use Seahorse
Connect RAG
Connect your existing vector database (Qdrant, Pinecone, Weaviate, ChromaDB). Seahorse creates its own metadata layer alongside your data. It never modifies your documents without your permission.
Upload PDFs
Drag-and-drop PDF files into the dashboard. Seahorse auto-chunks and embeds them. No vector database setup required. Start analyzing immediately.
Document Editor
Upload a single PDF, open it in a WYSIWYG editor with cognitive analysis as inline annotations. Review findings in context, edit the text, and export corrected documents.
Every few hours (or on demand), Seahorse runs a cognitive cycle: a multi-pass analysis that scans your documents, detects issues, audits observations for accuracy, evaluates health, learns new perception rules, and optionally applies fixes.
Core Concepts
Document Sources
Seahorse works with documents from any source. Connect a RAG for bulk analysis, upload PDFs for quick ingestion, or use the editor for focused single-document work. All sources feed into the same cognitive engine and can be searched together in the multi-source chat.
Cognitive Engine
9 specialized agents work together in an 8-phase loop. They scan your documents, analyze them through multiple "lenses" (orphans, contradictions, gaps, obsolescence), audit observations for accuracy, evaluate health metrics, generate perception rules, and apply fixes. The system literally reprograms how it perceives your knowledge base based on what it learns.
Learn more about the Cognitive Engine →
Observation Auditor
After lenses produce observations, the Auditor agent samples 30% and validates them against source documents using adversarial prompts. Observations marked "suspect" are filtered from metrics. Each lens accumulates a trust score based on its audit track record.
Multi-source Chat
Chat can search across PDFs, RAG, or both with a source toggle. Responses include clickable source citations with provenance (file name, page number). You can also generate documents directly from chat that open in the editor.
Supported Vector Databases
When using the Connect RAG path, Seahorse supports four providers:
Qdrant
Pinecone
Weaviate
ChromaDB
Get Started
Ready to bring your documents to life? Head to the Getting Started guide to create your account and run your first cognitive cycle.