A direct LLM prompt
Useful for a quick result
- One-time summaries
- Small document sets
- Exploratory analysis
- Disposable output
Open Knowledge Forge
Turn scanned documents into verifiable AI knowledge.
Open Knowledge Forge (okforge) transforms scanned PDFs and existing Markdown into structured, interlinked knowledge bases with page-level source citations—using models and infrastructure you control.
Explore a finished knowledge baseOpen source Local-first Plain Markdown MCP-enabled
More than a one-time prompt
Can’t an LLM create a knowledge base by itself?
Yes. A capable language model can turn a few documents into summaries or wiki pages with a well-designed prompt. For a small, one-time task, that may be enough.
But a prompt is an event. okforge is a repeatable system.
It preserves the connection between generated knowledge and original source pages, extracts visual content, applies a consistent structure across documents, and maintains an evolving knowledge base that can incorporate new sources over time.
okforge
Designed for durable knowledge
- Traceable Page-level citations connect knowledge to its sources.
- Consistent Documents follow a repeatable extraction and structuring process.
- Maintainable New material can be added to an existing knowledge base.
- Portable The result is stored as standard Markdown.
- Private The complete workflow can run on infrastructure you control.
Use a prompt when you need a quick result. Use okforge when the knowledge needs to be organized, verifiable, and built to last.
From source documents to structured knowledge
Extract. Forge. Explore.
Each stage can be inspected, configured, or used independently.
Extract
Turn scanned pages into structured Markdown.
Vision-language models transcribe text, identify visual content, and preserve the relationship between each output and its original page. Images and diagrams can be saved alongside the extracted text.
Forge
Transform sources into an interlinked knowledge base.
okforge identifies important concepts, creates summaries, connects related information, and generates knowledge pages with citations back to the original source pages.
Explore
Browse, search, or connect through MCP.
Use the knowledge base as a website, search it from the command line, or connect it to a compatible AI client that can read curated pages and trace their supporting sources.
Use the complete pipeline, or choose only the components your application needs.
Watch it work
See okforge in action.
Follow document extraction, Markdown generation, knowledge-base creation, source citations, browsing, and publishing in the working WebUI reference implementation.
Built for trusted, durable knowledge
Knowledge you can verify, control, and keep.
Verifiable
Trace every claim back to its source.
Page-level citations connect generated knowledge to the original document pages so people can inspect the evidence.
Local-first
Keep sensitive documents under your control.
Document processing, knowledge-base generation, and inference can all run on infrastructure you manage.
Portable
Own the knowledge, not just access to it.
Knowledge bases are stored as plain Markdown with structured metadata rather than locked in a proprietary platform.
Model-independent
Use the models and clients that fit your environment.
Keep factual knowledge separate from model weights and make it available through compatible tools and MCP clients.
See the result
Explore a finished okforge knowledge base.
The Dade County Building Code of 1935 was transformed from a scanned historical document into a searchable, interlinked knowledge base with concept pages, extracted source material, full-text search, graph navigation, and page-level citations.
The project was produced with SRI Consultants, a Southeast Florida engineering firm that uses historical building codes when researching older structures.
Produced with okforge in collaboration with SRI Consultants .
Knowledge page
Dade County Building Code of 1935
A structured collection of concepts, requirements, source extracts, and page-level citations.
Source: Dade County Building Code, page 42
Modular by design
Use the complete workflow—or only the parts you need.
Two core Python packages and an optional browser-based reference implementation can be used independently or together.
Document extraction
okforge-vision-ocr
Extract scanned documents into structured Markdown.
Transcribe scanned pages with vision-language models, extract images and diagrams, and preserve source-page mappings.
pip install okforge-vision-ocrKnowledge-base engine
okforge
Build, search, and serve knowledge bases.
Transform extracted or existing Markdown into an interlinked wiki with summaries, concepts, citations, command-line tools, and MCP access.
pip install okforgeReference implementation
okforge-webui
Run the complete workflow from a browser.
Combine extraction, Markdown generation, optional ingestion, knowledge-base management, browsing, and publishing.
View the projectBring your own AI client
Your knowledge base is not tied to one chat application.
okforge runs as an MCP server that allows compatible AI clients to search and read the generated knowledge base. It does not require a built-in chat interface or a vector database.
A connected model can search for relevant concepts, open complete curated pages, and follow citations back to the original source material.
Build the knowledge once. Use it with the compatible models and interfaces that fit your environment.
Open WebUI
llama.cpp
Page Assist
Hermes Agent
MCP server
okforge
Source-cited knowledge base
Markdown · concepts · citations
Your documents. Your infrastructure.
Keep processing and inference under your control.
okforge supports a fully self-hosted workflow in which source documents, generated knowledge bases, and model inference remain on infrastructure you manage. Hosted OpenAI-compatible endpoints are optional rather than required.
Document processing, knowledge-base management, and inference do not have to run on the same machine. A modest server can host the knowledge hub while inference runs on a compatible GPU system elsewhere on the local network.
Private by default
Keep sensitive source material and generated knowledge within systems governed by your organization.
Flexible deployment
Run okforge and model inference together or distribute them across separate machines.
No required cloud service
A fully local deployment needs no SaaS account or hosted service in the processing path.
Documents
okforge knowledge hub
Local inference server
Compatible AI clients
Start forging
Build knowledge you can inspect, move, and keep.
Use the complete browser workflow, install only the core components you need, or integrate okforge into your own application.
Prefer an assisted installation?
Give the prepared installation prompt to a compatible coding
agent for guided setup and configuration.
Open the installation prompt