AI mapping
The blank canvas problem, solved
Describe a topic or upload a document. Untangling returns a real map — nodes you can drag, rename, restyle and delete. Not a picture, not a wall of prose.
One sentence in, one map out
Untangling asks the model for structured JSON, validates it against a strict schema, repairs anything malformed, and only then builds the map. You never see a half-parsed response.
File to map
Feed it something you already have
A brief, a transcript, a chapter, a set of meeting notes. Untangling extracts the text, respects a size budget, and returns structure.
Reports, papers, briefs. Text is extracted page by page.
Word (.docx)
Documents keep their heading hierarchy where present.
Markdown
Headings and lists map directly onto branches.
Plain text
Indented outlines and transcripts both work.
Uploads are size-capped and MIME-validated. Text is truncated to a token budget before it reaches the model, so a 300-page PDF fails gracefully rather than silently costing you.
Per-branch actions
AI where you are stuck, not everywhere
Select a topic and choose an action. Results arrive as ordinary child topics you can edit, reorder or delete like anything else.
Expand ideas
Take a thin branch and grow it into three to six concrete sub-points.
Summarise
Collapse a sprawling branch into the two or three things that actually matter.
Generate subtopics
Ask for the next level down when you know the heading but not the detail.
Rewrite
Sharpen wording without changing meaning — useful before sharing.
Brainstorm alternatives
Get three other ways to frame the same branch.
How it works
Predictable, bounded, private
AI is a feature with edges, not a black box wrapped around your data.
Schema-validated
Model output is parsed with a strict schema. Invalid structure is repaired or rejected — never rendered half-broken.
Rate-limited and metered
Every request is rate-limited per account and recorded against your plan’s monthly credits, visible in Settings.
Not training data
We send only the text you submit, only when you ask. Your maps are never used to train a model.
Provider choice
Untangling talks to AI through a provider abstraction. Point it at Anthropic or OpenAI with an API key, or at your own compatible endpoint. Self-hosted deployments can run entirely inside their own network.
Start from something
Five AI credits are included free every month. That is five first drafts you did not have to write.
Create a free account