Show HN: ThoughtDAG – An editable context graph for LLM conversations
Points and comments are a snapshot, not live.
ThoughtDAG creates editable context graphs where edges control LLM context inclusion.
ThoughtDAG is a local-first, MIT-licensed tool that makes LLM context visible and editable via directed acyclic graphs. Each node is a prompt or response; edges determine which upstream nodes enter the next request. Users can delete an edge and regenerate to exclude branches without deleting history. Supports Ollama and OpenAI-compatible endpoints, PDF clipping with page provenance, and context preview with token counts. The interface is intentionally human-controlled, not automated retrieval.
What commenters are saying
Commenters broadly found the concept compelling but questioned the standalone app approach, suggesting plugin integrations for Claude Code, Codex, or editors instead. A security researcher identified a critical vulnerability from pdftoppm shell execution and a 0.0.0.0 bind; the author patched urgently, moving to 127.0.0.1 and removing shell execution. Some noted similarities to Obsidian canvas experiments and git branching, though the author emphasized the operational (not decorative) nature of edges. A minority criticized public security disclosure as irresponsible.