Build a self-organizing knowledge platform that answers questions with evidence
WeKnora consolidates scattered PDFs and wiki pages into a single system. It answers questions with cited evidence, executes complex tasks step-by-step via agents, and automatically organizes knowledge into wiki format. As an open-source LLM knowledge framework for enterprise document understanding, semantic search, and reasoning, it allows you to deploy on local or private cloud environments to keep data in your own infrastructure while handling everything from search to automation in one place.
v0.9.0: Multi-agent collaboration and automation
The recent v0.9.0 release introduces the WeKnora assistant, which explains release notes and settings and performs agent creation or task scheduling from the current page. A multi-agent collaboration feature has been added, allowing independent subtasks to be delegated to multiple agents, with progress tracking and result integration. Additionally, an automation feature has been introduced to handle one-time or recurring tasks with execution history and approval notifications, expanding the scope of agent operations.
Three modes: Search, Reasoning, and Wiki
WeKnora provides three modes on the same knowledge base. Quick Q&A (RAG) finds answers in the knowledge base and cites sources, while Smart Reasoning allows agents to plan and execute multi-step tasks such as searching, reading documents, and calling tools. The Wiki mode extracts people, products, and concepts from documents to organize them into interconnected Markdown pages and a knowledge graph, which can be edited directly in the browser with change history rollback capabilities.
Agent toolbox: Sandboxes and browsers
Agents execute skills installed from ClawHub, SkillHub, Git, or ZIP within Docker, E2B, or Cube sandboxes that maintain session state. An interactive terminal and graphical desktop open next to the chat, allowing real-time monitoring of the work process and direct intervention if needed. Through the BrowserSkill extension, agents can directly manipulate the user's Chrome or Edge browser to open pages or fill out forms, handing over to the user during authentication steps such as login or captcha.
Deployment and operations
WeKnora can be deployed via Docker Compose, Kubernetes (Helm), a single binary (Lite), or a desktop app. LLMs, vector databases, and storage backends are all swappable, enabling deployment on local or private clouds to keep data in your own environment. For operations, Langfuse tracks agent step-by-step reasoning, token usage, and pipelines, while a task queue dashboard manages the worker pool. Access is controlled through workspace-level RBAC (4-tier roles), audit logs, and scoped API keys.
Pre-adoption considerations
WeKnora is written in Go and requires Docker, Docker Compose, and Git. To use local Ollama models, you must first run ollama serve. For production deployments, it is recommended to deploy on internal networks rather than the public internet, configure firewall rules and access controls, and regularly update to the latest version for security patches. Environments deployed before v0.8.2 must check the upgrade notes, as the DingTalk channel has changed to Stream-only and sandbox commands are executed with root privileges.
