SourceWeft/SourceWeft

Open-source, self-hostable alternative to ChatGPT and NotebookLM — agents, skills, MCP, and cited knowledge in one AI workstation. Your models, your infrastructure. Free to use GPT 6: https://SourceWeft.com/

SourceWeft is an open-source AI workstation you can run on your own machine or your team's server. It's positioned as a self-hostable alternative to tools like ChatGPT and NotebookLM, letting you bring your own AI models, API keys, and data.

What it does

SourceWeft combines chat, knowledge, files, and task execution in one place. You describe a piece of work, and AI agents break it down, use the tools you've connected, and produce results you can review and refine. Conversations and files stay together, so each finished task becomes the starting point for the next one.

Key capabilities

  • Multi-agent collaboration — You can delegate research, exploration, planning, and execution to focused sub-agents that share working context and bring their findings together.
  • Skill library — Built-in skills cover presentations, HTML pages, images, meeting summaries, and learning. You can also add custom skills to reuse your own workflows.
  • Connected tools (MCP) — Agents can be extended with MCP servers, web search, and web extraction, letting them gather information and act on it.
  • Sandboxed execution — Approved commands and file processing run in an isolated, temporary environment, with outputs collected back into your working files. This requires a configured runtime and your approval.
  • Multiple models & BYOK — Choose supported model providers and bring your own API keys, configured separately for chat and indexing.
  • Cross-platform access — Use the web app or desktop clients for macOS (Apple Silicon), Windows (x64), and Linux (x64). You can access your deployed workstation remotely and return to saved conversations and files across devices.
  • Creation & working files — Produce reports, presentations, HTML pages, learning guides, audio overviews, and images, then iterate on them in the same place.
  • Knowledge with citations — Bring in PDFs, web pages, notes, YouTube, Notion, Google Drive, Gmail, Slack, and more. Agents can use these sources, and you can follow citations back to the original material.
  • Self-hosting & team collaboration — Deploy for yourself or your organization, manage your own database and file storage, and organize shared workspaces with roles and access controls.

Example use cases

  • Turn papers, web pages, and notes into research briefings with key claims, citations, and open questions.
  • Turn course material, videos, and PDFs into study guides, FAQs, and quiz drafts.
  • Turn product docs and team notes into reports, presentations, blog drafts, and launch materials.
  • Get traceable answers about decisions, projects, and shared context from connected team sources.

Getting started

You can download installers directly from the GitHub releases page for macOS (Apple Silicon), Windows (x64), or Linux (x64). For self-hosting, download a release bundle and follow the Docker installation guide — it runs the web app, API, background jobs, database, and file storage without needing Node or Rust installed on the host. After starting, open http://localhost:3000 and configure a model provider or bring your own keys.

A typical first task: create a workspace, add sources (upload files, paste URLs, write notes, or connect existing tools), then ask questions or create something — checking citations, letting agents use tools and skills, and refining results in saved conversations and files.

License

SourceWeft is licensed under the Apache License 2.0. Content under enterprise/, if present, is excluded and governed by a separate license. Billing is optional — the default core build runs without credits/pages billing; commercial subscription and payment UI are provided by the enterprise billing module.

Community

Bug reports, feature ideas, code, and design contributions are welcome via GitHub issues and pull requests. There's also a Discord server for community discussion.

Related

  • Project
    ScottRBK/forgetful

    Forgetful is an MCP server that provides a persistent, graph-based knowledge storage system for AI agents, allowing them to share and retrieve context across sessions using atomic, vector-embedded notes.

  • Project
    Fergana-Labs/stash

    Stash is an open‑source platform that captures AI‑agent runs, curates the resulting traces into durable knowledge, and lets you package that knowledge as reusable “skills”. It offers a Python SDK, REST API, CLI, and an optional reward‑model loop that turns human‑annotated preferences into prompt‑optimisation signals, enabling continuous improvement of agents without retraining the underlying model.

  • Project
    cv-superding/dsh-deepseek-web-login

    A DSH plugin that drives DeepSeek agents using your authenticated web session instead of an API key, allowing free access to advanced model features.

  • Project
    ttguy0707/CyberClaw

    A LangGraph-based general-purpose AI agent framework that implements a secure two-phase skill execution protocol and sub-agent delegation for observable and extensible workflows.