ThoughtDAG

What is ThoughtDAG

ThoughtDAG is a context graph tool for working with local agent conversations. It takes conversations across tools, indexes them, and lets you turn relevant turns into an editable graph rather than relying on a long, linear chat history. The product is built around a simple idea: the graph is the context.

Instead of treating chat as a single stream, ThoughtDAG walks the incoming edges for a node, orders the relevant ancestors, and builds the message sequence sent to the selected model. That makes the request structure visible before generation, not after the fact.

The site positions the app as a way to find the conversations that matter, then decide what the model sees next. It also shows that you can start from a code file, an exact phrase, a URL, or a paper.

Key features

• Editable context graphs: you can branch, prune, and merge conversation paths instead of losing them in a linear thread.
• Source-linked clipping: passages or figures can be turned into nodes while preserving provenance.
• Inspect before generation: the interface previews the source nodes, their order, and token count before the request is sent.
• Edge-based control: removing an edge changes the next prompt, so you can test how context affects the answer.
• Reproducible context: the same question can be asked again with a different context structure, making answer changes easier to trace.
• Token awareness: the product shows an incoming-context preview and mentions token differences when branches are removed.
• No hidden memory selector: the app emphasizes explicit context wiring rather than an opaque selection step.

How to use it

1. Open ThoughtDAG and start with a source such as a code file, a phrase, a URL, or a paper.
2. Clip the information that matters into a node, or ask directly from the source when that is more useful.
3. Connect nodes with wires so the relevant history becomes part of the graph.
4. Inspect the preview to see what the model will read, including order and token count.
5. Remove, keep, or merge branches, then send the request again to compare the result.

A practical workflow shown on the site is to isolate a research path, remove an unrelated branch, and regenerate the same prompt. The answer updates because the context changed.

Who it is for

ThoughtDAG is aimed at people who work with AI outputs but need tighter control over context.

• Researchers who want to keep evidence, interpretations, and detours separated.
• Developers who need a clearer view of which history is being passed to a model.
• Writers and analysts who clip passages and compare how different context changes an answer.
• Anyone using local agent conversations across tools and wanting a more inspectable workflow.

It is especially useful when the same prompt should produce different results depending on which prior turns, sources, or branches are included.

What to know before you use it

ThoughtDAG is not a general note-taking app or a simple transcript viewer. Its main job is to shape model context, not to archive every conversation detail for its own sake. The site does not present it as a full knowledge base, a citation manager, or a replacement for official source systems.

There are also practical limits to keep in mind:

• It can show and edit the context that gets assembled, but it does not claim to verify whether a source is true or complete.
• Token counts and previews describe what the selected model will receive, but they are still dependent on the current graph and selected nodes.
• If you need the authoritative record of a paper, code file, or source document, use the original source instead of relying on the clipped node.
• If your workflow depends on exact model behavior, test the same prompt more than once, because changing the graph changes the request.

The app is best treated as a way to make context explicit, inspectable, and reusable. If your work depends on knowing why a model answered a certain way, ThoughtDAG gives you a structured way to see and edit the input before it is generated again.