chenxiachan/thoughtdag
Branch AI conversations on an infinite canvas inside DeepSeek Harness: each exchange is a node and the wires between nodes are the context the model receives next.
ThoughtDAG turns an AI conversation into a branching graph on an infinite canvas: each exchange becomes a node and the wires you draw between nodes are the context the model receives next, so you can explore a side question and connect only the paths you want without deleting work. Inside DeepSeek Harness you switch between chat and the graph, choose the context on the canvas and the harness runs the next turn. The 0.5 line adds a local history index (searches supported agent sessions and ThoughtDAG canvases, builds topic dossiers) and an optional Jev decision layer that helps identify topics and rank relevant excerpts — the README reports a small relevance-selection pilot where Jev's median selection was 391 ms versus 24,813 ms for its GLM adapter, while stressing these are selection-stage timings, not end-to-end search or answer times.
Install
dsh plugin --profile web add dsh-thoughtdagCopied verbatim from the README's 'Inside DeepSeek Harness' section, which for the web profile gives these two lines (dsh plugin --profile web add dsh-thoughtdag then dsh web). The README also describes the desktop-app path (Plugins → search dsh-thoughtdag → install → restart Harness) and warns that a release published within the last 24 hours needs its version named, dsh-thoughtdag@<version>, because pnpm holds such versions back by default.
Compatibility
The README's plugin section states the plugin bundles the canvas and memory layer and requires Node 22.19+ (22.x) or 24+, plus DeepSeek Harness 0.1.2-rc.1 or later. The desktop app is a separate brew install --cask thoughtdag path, and the CLI half is npm install -g thoughtdag with thoughtdag setup mcp to expose read-only history tools.
Details
- Repo: chenxiachan/thoughtdag
- Category: Sessions & Messages
- Stars: 574
- Version: GitHub repo chenxiachan/thoughtdag (MIT, per the README's license badge); npm package dsh-thoughtdag at 0.5.17 (registry check 2026-10-07). The README states it requires Node 22.19+ (22.x) or 24+, and DeepSeek Harness 0.1.2-rc.1 or later.
FAQ
- How does the graph change what the model sees?
- The README's one rule is 'wires are the context' — connected paths are used in the next question — and you can preview what the model will receive before sending.
- Is the past-conversation recall automatic?
- It is opt-in: with recall enabled the context panel lists the dossiers and excerpts added to a request so you can inspect their sources or exclude individual items before continuing.
Alternatives
LeslieWylie/dsh-session-search-pro · beijingwahw/dsh-companion · PwnKY/dsh-session-link