Knowl
Your project knowledge, maintained as a wiki.
Knowl
baldaworks4 Sterne
Self-hosted LLM wiki turning project context into OKF memory
- Transport
- stdio
- Status
- Lokal
Agenten stärken
Installation vom Herausgeber. Kein PlugMCP-Endpunkt.
Host a copy with our keys and logs.
Noch kein paketiertes Release in der Registry. Folgen Sie der README des Repositorys.
Remote-URL oder ein Befehl auf Ihrem Rechner? Worin sich die beiden Installationen unterscheiden
Auf PlugMCP ausführen
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
StartenKnowl is a self-hosted LLM wiki for your projects. It turns source documents into connected Markdown pages and gives AI agents grounded evidence from that wiki. Knowledge remains available across tasks, with references back to the sources that support it.
Knowl builds on Andrej Karpathy's LLM Wiki pattern: an agent reads sources and maintains connected knowledge over time. Bring your own ACP-compatible agent to maintain the wiki. Your knowledge stays in portable, Open Knowledge Format (OKF) Markdown under your control.
Connect project sources or start from an existing Markdown wiki, Obsidian vault, or OKF bundle. Knowl preserves accepted source revisions, and the configured LLM maintainer proposes semantic updates to the wiki. Updates pass validation before they become part of the knowledge base.
A URL submission stores a reference. To use the page's contents as a source, submit its text. See the ingest examples.
Your agent retrieves relevant evidence and uses it to answer or act. You can also read the wiki directly, inspect its source references, or export it for publishing. See the source-to-wiki example for a complete walkthrough.
Both guides take you from a source document to a published wiki page and a
retrieval result. Choose your maintainer in the agent guide,
including a custom agent through generic_acp over stdio.
The container quickstart is now in the service guide.
Knowl maintains a wiki of its own documentation. Compare the source docs with the generated wiki to see source material become connected knowledge. The walkthrough explains the configured ACP agent, generation commands and saved source provenance.
Use Knowl from a local Codex project with setup and maintenance skills plus MCP tools. For the v0.8.0 npm release, setup is:
npx --yes @baldaworks/[email protected] setup codex
The npm instructions require that release to be published.
After setup, start a new Codex thread and use $knowl:run for one bounded
wiki maintenance cycle.
See the local Codex guide for requirements, provider
configuration, offline use, and supported scope.
Connect other agents to the self-hosted service over MCP. Knowl exposes three
tools: knowl_retrieve for evidence, knowl_ingest for durable inputs, and
knowl_operation for operation status.
The same capabilities are available through HTTP. See service connection guide, service operations, and the OpenAPI contract.
Knowl can request one corrected plan when generated output fails validation. Inspect the result through operation status, or disable correction in your config.
Embed Knowl through pkg/knowl, or use pkg/knowlfx for Fx lifecycle
integration. See the product design for runtime composition
and ownership boundaries.
Enable the optional web UI to browse published pages, inspect their saved source revisions, retrieve evidence, and follow maintenance status. It runs inside the Go service with locally bundled assets and is disabled by default. Connect with an operator token; use HTTPS when exposing the service beyond loopback.
Enable optional embeddings to retrieve related evidence across paraphrases and word forms. Knowl combines semantic and keyword search while keeping the original page text and source references in its results. The semantic index covers the full text of each published page within the configured projection capacity; a dense-only match shows the matching passage as its excerpt.
Use your own compatible embedding API or the local CPU embedding server. Dense scoring runs natively inside SQLite or optional operator-provisioned pgvector; see deployment/rebuild guidance and the actual performance comparison. Zero stored-vector transfer does not promise faster queries: current paired measurements remain mixed, and the original full-query regressions are retained. Full integrity is established before atomic publication; ordinary reads check scalar readiness, snapshot/model compatibility and consumed evidence. Lexical-only remains the default. Embeddings are off by default. Choose explicit keyword fallback or strict semantic availability in the search guide.
The canonical wiki lives in wiki/ as an OKF v0.2 bundle. Accepted source
revisions are preserved separately in raw/, so semantic pages remain connected
to their evidence. The default local operational store uses SQLite.
Export the wiki as a standalone bundle, or add an llms.txt navigation file for
agents and publication. See workspace and export for the
layout and commands.
See the documentation index for architecture, API contracts, quality measurements and release history.
Knowl is released under the MIT License.
Self-hosted LLM wiki turning project context into OKF memory
Tools
Noch keine Tool-Liste. Remote-Server erhalten eine nach dem ersten Health-Check.
FAQ
Kopieren Sie auf dieser Seite die stdio-Konfiguration für Cursor, Claude Desktop, Claude Code oder VS Code. Sie nutzt die Installation des Herausgebers, keinen PlugMCP-Proxy.
In diesem Eintrag ist keine erforderliche Umgebungsvariable aufgeführt.
Knowl läuft lokal über stdio.
Zuletzt geprüft: 2026-11-12. Self-hosted LLM wiki turning project context into OKF memory