An open-source Human+AI collaborative knowledge platform built on the W3C Web Annotation standard, enabling document-grounded knowledge graph construction and semantic retrieval through seven composable flows.
Semiont is an open-source semantic wiki system developed by The AI Alliance, built on the W3C Web Annotation standard as its data foundation. All knowledge is expressed as structured annotations anchored to specific passages in source documents. Its core design philosophy is "document-grounded knowledge graph" — every node in the graph can be traced back to an original document fragment rather than existing as a decontextualized abstract entity.
The system defines seven composable collaborative flows: Yield (resource ingestion), Mark (structured annotation), Match (multi-source retrieval with ranking), Bind (entity disambiguation and linking), Gather (context assembly), Browse (structured navigation), and Beckon (attention guidance). Human users and AI Agents are architectural peers, sharing the same API, event bus, and event-sourced storage — all operations produce events in a uniform format.
The technical architecture follows a Spec-first approach where OpenAPI specifications precede implementation and the TypeScript SDK is auto-generated. Core storage is based on event sourcing. The monorepo structure splits into an application layer (Hono backend, Vite+React frontend, Node.js CLI) and a core packages layer (api-client, event-sourcing, content, graph, ontology, inference, make-meaning, jobs, mcp-server, etc.). Inference backends support hybrid deployment of Anthropic (cloud) and Ollama (local), configurable at Worker granularity. An MCP Server and Claude Code Agent Skills definitions are provided for direct integration into AI assistant workflows.
Use cases include academic literature annotation and citation tracking, legal and compliance document cross-review, organizational knowledge management, RAG grounded in annotated source materials, and providing AI Agents with persistent structured memory. The project is in active Alpha stage (v0.4.22) with unstable APIs that may include breaking changes. Primary language is TypeScript (90.3%), supporting Web browser, desktop (macOS/Linux), and CLI interactions. Deployment is containerized with frontend images published on GHCR, supporting Docker/Podman/Apple Container.