Build real-time agents that react to new information
Subscribe to any fact and get notified when it changes.
openFDAMounjaro label
ClinicalTrialsSURPASS-CVOT
EMAWegovy EPAR
Powered by 50+ data providers
How it works
One call to create a watch. We keep checking the sources and tell your agent when something changes.
Create a watch
One call with a query or a claim. Add a webhook URL if your agent runs as a service.
We keep checking
A trial registry, a drug label and a guideline change at different speeds. The provider graph refreshes each source at its own pace.
Get the change
A deployed agent gets a webhook with the before, the after and the sources. A harness like Claude Code reads the changes at the start of its next session.
Watch, from your terminal
Create a watch, then read what changed since your last session. Your coding agent gets the same calls through the KIRHA skill.
Works in any harness
Get your API key, then install with one command
- π
More than Search on a schedule
You could run a search every morning and compare the output. Watch does the parts a scheduled job cannot.
Changed facts only
Events list the facts that were added, removed or changed. A reworded page footer does not wake your agent.
Watch a claim
Give us a statement and every run checks it again. The watch fires when its verdict changes, for example from supported to contradicted.
A history nobody can backfill
Every change is stored with its date and source: when a label changed, which warning was added, how long a trial stayed recruiting. A change missed on the day is gone for anyone who starts watching later.
Evals that know the date
Each change carries the before and the after. Build evals against the label as it stood, and re-run them when it changes.
Know the day the data changes
Point a watch at the labels, registries and trials your agents depend on. Your agent hears about the change the day it happens.