Documentation
Everything you need to run open-core Kvendra yourself and understand how it works. New here? Start with the Quickstart. For the big picture first, read How it works.
Start here
- Quickstart — The self-hosted golden path: reference stack, Skills, connect Claude Code, onboard your project.
- Install with your AI — Let Claude Code read llms.txt and run the Quickstart for you.
- Advanced install — Step by step: engine only, your own Postgres, build from source, every setting.
KB Engine (the required core)
The self-hostable knowledge base your agent reads and writes — start here.
- Platform — A self-hostable KB on PostgreSQL that finds your project knowledge by meaning.
- Entities — The typed entity model: what each type is for and how they relate.
- Governance — Guarded updates, proposals and approvals, decision keys, and imported content as data.
- Session brief — Resume where you left off with whoami, see peers with the radar, pull changes with check_notifications.
- Private values — Keep identifying values out of KB text: project private values and per-machine local variables.
- Workspace files — Binary attachments for entities: upload flow, references, limits and roles.
Skills (your day-to-day)
The Claude Code plugin that runs pipelines as one-command workflows on the Platform.
- Skills — The Claude Code plugin: thin skills, with the knowledge living in the KB.
Pipelines
How changes land together, or not at all.
- Pipelines — Transactions, draft/activate/cancel, and the typed server-side changelog.
- Pipeline autonomy — How many human gates the orchestrators keep: dual, single or zero-gate with a hard floor.
CLI & Security (optional to get started)
An optional security layer you add when you want your agent to run real ops without putting your tokens in the model's context by default.
Concepts
- How it works — How the Platform, the Skills and the CLI are wired together over MCP.
- Principles — The design stances behind the project.
- Use cases — What you get out of it, day to day.