Show HN: ThoughtDAG – An editable context graph for LLM conversations

Aug 15, 2026 11:42 AM - 3 hours ago 1

CHAT HIDES CONTEXT.

THE GRAPH IS THE CONTEXT.

Linear conversation

Editable discourse graph

Same punctual · different context

AI conversation 87 messages

Compare 3 investigation paths.

Start pinch the first. Its advantage is…

What if the halfway presumption fails?

Consider different explanation…

Also, what should I eat tonight?

There is simply a caller edifice nearby.

The history is here. Which parts participate the adjacent request?

research-paper.pdfp.7

Results

The effect appears only successful the experimental condition.

Selected from the page

Clipped passageresearch-paper.pdf · p.7

The effect appears only successful the experimental condition

Source linked · not wired yet

Asked from sourceresearch-paper.pdf · p.7

What does this grounds really mean?

The root is successful context

Unrelated branchdetour

What should I eat tonight?

This history should not participate the investigation summary.

Still connected

Polluted summary 3 sources

Research summary… also, see basking cookware for dinner.

The punctual stayed the same. Polluted discourse changed the answer.

Includes unrelated branch

Will send1,284 tokens

Preview what the exemplary will receive

Incoming ancestors:

Research question Evidence A Dinner detour

After deleting the orangish edge: −47 tokens

Context diff−47 tok

The meal detour near context

Same punctual · regenerate

Same promptask again

Give maine a bullet-point summary

The words are identical. Only 1 separator changed.

Reproducible context

Clean answer 2 sources

One: grounds the database version. Two: usage independent reviewers. Three: resoluteness conflicts pinch a 3rd reviewer.

The unrelated meal proposal is gone.

Answer updated successful place

One ruleThoughtDAG

Wires are context.

No hidden representation selector. What the exemplary sees, why, and what was removed enactment visible successful the graph.

Visible Editable Inspectable

01 · The problem

Chat history is long. Context is still invisible.

The interface shows what was said, not which history enters the adjacent request.

02 · Externalize

Ask from the source. Clip what matters.

Ask from a selected passage, aliases move a transition aliases fig into its own source-linked node. Provenance stays attached; discourse remains yours to wire.

03 · Inspect

Before sending, inspect what the exemplary will read.

Preview root nodes, order, and token count. Context is nary longer a hidden decision.

04 · Edit

Delete 1 edge. Ask the aforesaid mobility again.

The removed branch really leaves the request. The reply changes pinch the context.

05 · The protocol

Most canvases shape information. ThoughtDAG edits context.

You determine what enters and leaves. The chart is the discourse protocol before generation.

1 / 5 Invisible context

More