- The SDK in your product — identifies your users, tracks their behavior, and renders the in-app messages they’re eligible for.
- The skill in your agent — teaches Claude Code to create, launch, and measure communications through the management API.
Get your keys
Every project has two keys, both under Settings → Projects in the dashboard:- the publishable key (
pk_pub_…) — for the SDK; safe to ship in the browser, - the secret key (
pk_…) — for your agent and backend; never expose it to a browser.
Signups are currently limited while Galinum is in early access. The
dashboard shows your identified users, tracked events, campaign performance,
and your agent’s run history as soon as data flows.
Install the SDK
You need a React 18+ app (any framework; in the Next.js App Router the SDK renders inside Client Components).@galinum/react ships as a tarball in each Galinum product release. Install
it from the file:
The SDK version matches the product release it ships with.
Wrap your app
Identify your users
WithuserId set, the SDK identifies the user (attaching
auto-collected context)
and fetches the messages they are eligible for. <InAppMessages/> shows one
per page view —
a default card out of the box, or your own UI.
You can also identify explicitly, with traits your agent can target on:
reset() so tracking and message delivery stop until the next
identify.
Track your first event
Behavioral events are what your agent analyzes to choose an audience — “free users who haven’t tried CSV export” is a trait plus anot_seen event:
Install the skill in your agent
Copy the galinum skill into your repo at.claude/skills/galinum/ and make the project’s secret key available to
your agent (for example as GALINUM_SECRET_KEY in its environment). Any
Claude Code session in that repo can now operate Galinum.
Ship your first communication
Tell your agent what you want to announce:<InAppMessages/> on their next page view, and
impressions, clicks, and dismissals show up on the campaign in the dashboard.
That’s a direct communication — no goal, no
A/B test, done once it’s live. Ask for more when you want it: a delivery
window (“run it for a week starting Monday”), an outcome check (“tell me
whether people start using it”), or goal-driven optimization (“A/B test this
against the activation goal”).
How Galinum works
Organizations, projects, keys, and the communication model.
React SDK docs
Custom rendering, auto-context, and the full API surface.
Claude Code skill
What the skill does and the full skill source.
HTTP API reference
Every SDK and management endpoint, for custom clients and agents.