pref0 vs Retrieval-Augmented Generation (RAG)

RAG retrieves documents to augment LLM responses. pref0 retrieves learned preferences. They solve different problems and work well together.

pref0RAG
What it retrievesStructured user preferencesDocuments and text chunks
Data sourceExtracted from conversationsPre-indexed document corpus
PersonalizationPer-user preference profilesSame documents for all users
Learns over timeYes — confidence compoundsNo — documents are static
InfrastructureHosted API, no vector DBVector database + embedding pipeline
Best forHow the user wants things doneWhat the agent needs to know

Key differences

Preferences vs. knowledge

RAG provides knowledge — facts, documentation, data the LLM doesn't have. pref0 provides preferences — how the user wants the LLM to behave. These are orthogonal concerns.

Per-user vs. global

RAG typically retrieves the same documents regardless of who's asking. pref0 retrieves a different preference profile for each user. Both are injected into the prompt, but they serve different purposes.

Static vs. learning

RAG documents are indexed once and updated manually. pref0 preferences learn and compound automatically from every conversation. The agent gets more personalized over time without manual intervention.

When to use each

Use pref0 when...

  • You want to personalize how the agent responds to each user
  • You need the agent to learn from corrections and feedback
  • You want confidence-scored preferences, not document retrieval
  • Your agent is already good at its task but needs personalization

Use RAG when...

  • You need to ground the agent in specific documentation
  • You want to augment the LLM's knowledge with your data
  • You need the agent to cite specific sources
  • Your primary challenge is knowledge, not personalization

Frequently asked questions

Should I use pref0 instead of RAG?

They solve different problems. RAG gives your agent knowledge. pref0 teaches your agent preferences. Most production agents benefit from both — RAG for what to say, pref0 for how to say it.

Can I use pref0 alongside my RAG pipeline?

Yes. Inject pref0 preferences into the system prompt alongside your RAG context. The agent gets both knowledge and personalization.

Does pref0 use vector search?

No. pref0 returns a user's full preference profile directly. No embedding, no vector DB, no similarity search. It's a simpler architecture.

Other comparisons

vs Mem0vs Zepvs Custom Memoryvs LangChain Memoryvs LangMemvs Lettavs ChatGPT Memoryvs Pineconevs CrewAI Memoryvs Semantic Kernel

Not memory. Preference learning.

Your users are already teaching your agent what they want. pref0 makes sure the lesson sticks.

Get Started FreeView Docs