TokenPath Docs

TokenPath is source attribution for AI applications. You send three strings:

document: the source text your model read,
question: the question or task you gave it,
answer: the answer it wrote.

TokenPath returns links from phrases in the answer to the exact text in the document that supports them, each with a confidence score. Your app turns those links into citations, highlights, or click-to-source review.

{
  "spans": [
    {
      "answer": { "start": 22, "end": 38, "text": "green and yellow" },
      "source": { "start": 115, "end": 131, "text": "green and yellow", "confidence": 0.704467 }
    }
  ],
  "offset_encoding": "utf-16"
}

TokenPath works with any model. It does not generate, rewrite, or sit in your inference path. It runs after the answer exists.

Quickstart

Make your first request in two minutes. No key needed.

Set up with a coding agent

Point Claude Code or another agent at skill.md and it wires TokenPath into your app.

Inline citations

Turn answer-to-source links into numbered citations.

PDF attribution

Highlight the exact supporting text on the PDF page.

API reference

Every endpoint, field, and error code.

Which endpoint should I use?

Endpoint	Use it for
`POST /v1/attributions`	Citations, source highlights, reviewable summaries, and almost every product integration. Start here.
`POST /v1/attributions/heatmap`	The raw token-by-token heatmap, for research, debugging, and custom visualizations.

Get started

Core concepts

Advanced

Use cases

Cookbooks

Resources

Quickstart

Set up with a coding agent

Inline citations

PDF attribution

API reference

Which endpoint should I use?

Quickstart

Set up with a coding agent

Inline citations

PDF attribution

API reference

​Which endpoint should I use?

Which endpoint should I use?