semiont
Semiont
Semiont is an open, source-grounded platform for building trusted AI knowledge bases — a shared workspace where humans and AI agents annotate, connect, and govern domain knowledge.
- Annotate and link documents. Humans and AI agents mark entities, comments, and references directly in your corpus — every annotation anchors to a specific passage, in the W3C Web Annotation standard.
- Grow a grounded knowledge graph. Annotations and links project into a graph where every node traces back to its source. Search it semantically, navigate it, audit its provenance.
- Serve trusted context to AI. Feed RAG pipelines, agents, and applications from cited sources instead of unchecked generation. Self-hosted, with inference on Anthropic (cloud) or Ollama (fully local).
No cold start. Most knowledge systems are useless until someone invests weeks in schema design, taxonomy building, and manual data entry — the cold-start problem. Semiont skips it: import documents and AI agents immediately begin detecting entities, proposing annotations, and linking related material for humans to review and refine. The knowledge graph grows as a byproduct of that work — no upfront schema, no ETL pipeline.
Get Started
Three steps — no npm or Node.js required. Install the semiont launcher (a single static binary) and a container runtime (Apple Container, Docker, or Podman):
brew install the-ai-alliance/semiont/semiont
Clone a knowledge-base repo — the empty template for a new project (or birth one in place with semiont init), or a pre-populated demo — and start it:
git clone https://github.com/The-AI-Alliance/semiont-gutenberg-kb.git
cd semiont-gutenberg-kb
semiont start
One command brings up the whole stack from published, attested container images — including the Semiont browser. Create your admin user and sign in at http://localhost:3000:
semiont useradd --email admin@example.com --admin # prompts for the password
Explore the knowledge bases:
- semiont-template-kb — Empty template; start here for a new project
- semiont-gutenberg-kb — Public-domain literature from Project Gutenberg
- Full catalog — seven demo KBs across different domains, plus community knowledge bases
See the Quick Start for full setup instructions.
How it works
Humans and AI agents are architectural equals: every operation — whether it comes from the GUI, the TypeScript SDK, agent skills, or the semiont launcher — travels the same event bus, speaking the same eight verbs: browse, bind, yield, mark, frame, gather, match, beckon. Any workflow can be done manually, automated by an agent, or shared between the two. The protocol docs cover the design in depth.
Open Source & Community
Semiont is Apache 2.0 licensed and developed in the open. We welcome contributions from the community.
- View on GitHub — Explore the source code and documentation
Part of the AI Alliance — building open, safe, and beneficial AI for everyone.