Research papers, structured for humans and machines
Every AI agent reading a PDF re-extracts the same claims from unstructured prose. Knows.Academy solves this — structure the paper once, consume it everywhere.
📄
paper.pdf
- ✗ ~10,000 tokens per read
- ✗ Every agent re-parses
- ✗ No structure, no IDs
- ✗ Can't verify claims
→
AI extracts once
⚡
paper.knows.yaml
- ✓ ~4,500 tokens (55% less)
- ✓ Parse once, fetch parts
- ✓ Every claim has an ID
- ✓ Confidence + provenance
What's in a Sidecar
Claims
What the paper asserts
Methods
How the work was done
Evidence
Results and measurements
Relations
How entities connect
Artifacts
Papers, code, datasets
Provenance
Who, how, when
Two Interfaces, Same Knowledge
🧑 For Humans
- • Summary + key claims at a glance
- • Interactive knowledge graph
- • Click any claim to see supporting evidence
- • Browse by discipline, search by content
🤖 For Agents
- • Schema-validated JSON via API
- • Partial fetch: only get claims, evidence, or citations
- • 55-99% fewer tokens than reading PDFs
- • BibTeX citations to reduce hallucination
Install Knows Skill
Add the Knows skill to Claude Code or any LLM agent. The skill auto-configures the API endpoint based on where you install from.
Copy this prompt and paste it to any AI agent (Claude Code, OpenClaw, etc.). The agent will download and register the skill automatically.
agent install prompt
The skill file is always the latest version. Re-run to update.
0 papers already structured.
Same knowledge. Two interfaces. Zero re-parsing.