chenxiachan/thoughtdag

Your thinking deserves a map: an infinite canvas where LLM conversations grow into an editable thought graph. Wires are the context.

What it solves

ThoughtDAG addresses the problem of linear chat histories and automatic RAG systems where irrelevant or incorrect information (noise) persists in the model's context, leading to hallucinations and wasted tokens. It allows users to visually manage and prune the context that a language model sees for any given turn, ensuring only relevant information is passed forward.

How it works

The project implements a Directed Acyclic Graph (DAG) where nodes represent messages, materials, or conclusions, and wires (edges) explicitly define the context. A model's input for a specific node is determined solely by the paths that wire into it. Users can delete edges to remove noise or merge multiple paths of thought into a single conclusion.

Who it’s for

Researchers, developers, and power users who need to maintain high precision in LLM interactions, manage complex investigations across multiple agent sessions (e.g., Claude Code, Codex, DeepSeek Harness), and visually map out their thinking process.

Highlights

  • Editable Context Canvas: Visually prune or merge conversational branches to change model output.
  • Session Atlas: Integrates and indexes conversations from various local agents into a single editable graph.
  • Clipped Material Integration: Read PDFs or files and ask questions directly from selected passages, with provenance tracking.
  • Context Search CLI: A command-line tool to find specific code mentions or phrases across all local agent sessions.
  • DeepSeek Harness Plugin: Integrates directly into the DeepSeek Harness web UI as a visual context switch.
  • Local-first Privacy: Can run fully offline via Ollama, keeping keys and documents on the user's machine.

Related

  • Dispatch
    ThoughtDAG: Managing LLM Context via Editable Graphs

    ThoughtDAG is an open-source, local-first tool that replaces linear chat histories with an editable context graph, allowing users to explicitly control which previous messages and documents enter the LLM prompt.

  • Project
    rahulnyk/knowledge_graph

    A tool that converts text corpora into interactive knowledge graphs of concepts using local LLMs, enabling advanced text analysis and Graph RAG.

  • Project
    NVIDIA-NeMo/Guardrails

    An open-source toolkit for adding programmable guardrails to LLM-based conversational applications to ensure safety, security, and controllable dialogue flows.

  • Project
    microsoft/graphrag

    A data pipeline and transformation suite that uses knowledge graphs to extract structured data from unstructured text, enhancing an LLM's ability to reason over private data.

  • Project
    juggler-ai/juggler

    A visual workbench for AI coding agents that replaces linear chat with a tree-structured interface for managing context, inspecting model transactions, and controlling agents remotely.