pref0 vs Letta (formerly MemGPT)

Letta is a stateful agent platform where agents manage their own memory. pref0 is a focused preference learning API. They operate at different levels of abstraction.

pref0Letta
What it isPreference learning APIStateful agent platform
What it storesStructured preferences with confidence scoresAgent-managed memory blocks (persona, user info, archival)
How memory worksAutomatic extraction from conversationsAgent self-edits its own memory via tool calls
Confidence scoringBuilt-in, compounds over timeNot available
ScopeAdd preference learning to any agentFull agent framework with built-in memory
Best forLearning user preferences across sessionsBuilding stateful agents from scratch

Key differences

API vs. platform

pref0 is a lightweight API you add to your existing agent. Letta is an entire agent platform — you build your agent within Letta's framework. If you already have an agent, pref0 adds preference learning without changing your architecture.

Automatic extraction vs. self-editing

pref0 automatically extracts structured preferences from conversations — no agent logic needed. Letta agents manage their own memory using tool calls, deciding what to remember. This is powerful but adds complexity.

Preference focus vs. general memory

pref0 is purpose-built for preferences with confidence scoring. Letta provides general-purpose memory blocks. If you specifically need 'user prefers X with confidence Y,' pref0 gives you that directly.

When to use each

Use pref0 when...

  • You have an existing agent and want to add preference learning
  • You want structured preferences with confidence scores
  • You prefer a simple API over adopting a full platform
  • You don't want to rewrite your agent in a new framework
  • You need a minimal integration (2 endpoints)

Use Letta when...

  • You're building a new agent from scratch and want built-in memory
  • You want agents that manage their own context window
  • You need the MemGPT architecture (self-editing memory)
  • You want a complete agent platform, not just a preference layer

Frequently asked questions

Can I use pref0 with Letta agents?

Yes. Add pref0 preference data to your Letta agent's system prompt or persona block. Use Letta for general agent state and pref0 specifically for structured preference learning.

Is Letta overkill if I just need preferences?

If preference learning is your main goal, yes. Letta is a full agent platform. pref0 is 2 API endpoints focused on preferences. Use pref0 for preferences and your existing agent framework for everything else.

How does pricing compare?

pref0 offers 100 free requests/month, then $5/1,000 requests. Letta has a free tier (3 agents), Pro at $20/month, and Max at $200/month. pref0 is simpler pay-per-use pricing.

Other comparisons

vs Mem0vs Zepvs Custom Memoryvs RAGvs LangChain Memoryvs LangMemvs 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