AI agents · Second message
Apache Unomi 3 for AI agents
Privacy-first customer memory — retrieve, store, summarize, and build segments. Unomi is not an LLM; it is the data layer agents plug into.
Primary site story remains Introducing Apache Unomi 3. This post is the strong second message for AI builders and SEO/AEO.
AI agents are only as useful as the customer context they can see — and only as trustworthy as the consent behind that data. Apache Unomi 3 gives agents a real, queryable, updatable customer memory over REST/JSON (and GraphQL).
1. Retrieve
Pull profiles, recent events, segments, scores, and personalization outcomes to ground copilots and personalize replies.
2. Store
Write preferences, intent, and conversation-derived facts as events and/or profile properties so agent memory is inspectable and governed like any other CDP data.
3. Summarize
Process history into profile briefs or enrichment fields — synchronously via agents, or asynchronously with Unomi’s task scheduler.
4. Operate
Propose or create segments, adjust scores, suggest rules, and drive personalization — agents can help operate the CDP, not only read it.
Why Unomi 3 for AI builders
- Stable, additive REST (119/119 documented 2.x paths retained) + optional V2 compatibility
- Multi-tenancy & API keys for SaaS-style agent platforms
- Elasticsearch or OpenSearch
- Cluster-aware tasks for enrichment jobs
- Open source and model-vendor-neutral — bring your own LLMs / MCP tools