AI infrastructure that keeps the evidence behind health claims current as research changes.
Built for the teams who substantiate claims and for AI agents.
Health evidence keeps moving
New trials appear, studies get retracted, and a claim that held last year may not hold today. Yet evidence is assessed once and left to age.
AI agents answer health questions at massive scale, and they get the evidence wrong, even with web search:
- Hallucinated citations to studies that don't exist
- Confident answers when the evidence is thin
- Hardly ever a "we don't know"
Grounding that shows its work
OpenFrames is the evidence layer for health AI. It finds the studies behind every claim, grades them and keeps them current.
One search, then a black box
- One semantic search per question
- No record of what was retrieved
- No way to tell if what came back was useful
- Nothing kept for the next AI query
Grounding you can audit
- An exact record of every study retrieved
- Which evidence the agent used for which part of its answer
- Scores and analysis for every study against the claim
- Kept, dated and updated when research changes
A Frame
OpenFrames builds a shareable, reusable object for every health claim: the studies for and against it, graded, with an advisory status and a dated history.
Claim in, Frame out
Parse
Splits the claim into intervention, outcome, effect and population.
Retrieve
Searches medical literature, clinical trials and other specialized sources for studies on it.
Grade
Rates each study for design, fit to the claim and reliability, and marks it as supporting or challenging.
Record
Stores the Frame with a date and a stable URL, and logs every change.
Watch
Flags the Frames a new study or retraction affects.
AI agents
Call the API or MCP and cite the Frame in every answer.
Review teams
Build and maintain claim dossiers on graded, dated evidence.
Built to be checked
Every citation resolves
OpenFrames only cites what a search returned.
"Insufficient evidence" is an answer
When the studies are thin, the Frame says so.
The status is advisory
We grade the evidence. Your reviewer, or your agent's policy, makes the call.
Every assessment has a date
You can see what the evidence looked like when a claim was checked.
Give your agent something to cite
Call the API, or connect the MCP server and let your agent request evidence on its own. Each Frame lives at a stable URL on frames.sh, so every answer can link to its sources.
pip install openframes
from openframes import Client evidence = Client().evidence("Magnesium improves sleep quality in adults") print(evidence.status, evidence.url)
Who it's for
Substantiation specialists
Consultancies that build claim dossiers for supplement and health brands spend weeks finding, reading and grading studies, and a dossier starts aging the day it ships. OpenFrames gives your team a graded starting point for every claim and a dated record of how the evidence was assessed. It is designed around the FTC standard of competent and reliable scientific evidence.
- Start each dossier from a graded evidence set
- Check a client's claims against the current literature
- Keep a dated record for every claim you sign off
- Get flagged when new research affects a client's claim
AI agents
Health agents answer millions of questions a day and need evidence they can stand behind. OpenFrames gives them graded studies, a status that says when the evidence is thin, and a stable URL to cite.
- Ground answers in graded studies
- Say "insufficient evidence" when the studies are thin
- Link every answer to its evidence
- Connect over API or MCP
About us
We are a team of AI, NLP and ML engineers. We have built language technology together for more than ten years. That includes text pipelines over tens of millions of documents, claim checking against web evidence, and a biomedical claim engine that grades studies from specialized sources.