Sonzai

What is Sonzai?

Sonzai is the relationship layer for consumer AI agents — a managed platform that gives AI companions, characters, avatars, and NPCs persistent memory, personality, mood, relationship state, and model-agnostic orchestration through one API. Build with our SDK, or bring in our forward-deployed engineers to integrate the relational harness directly into your product.

Who is Sonzai built for?

Sonzai is built for teams shipping consumer-facing relational AI: AI companion apps, AI character platforms, creator and celebrity avatars, game NPCs, social agents, and branded consumer AI personalities. The common thread is high-frequency user-agent interaction where retention depends on the agent actually knowing the user.

What does a relationship layer give my agent?

Persistent memory across sessions, a consistent personality, a current emotional mood, a relationship arc that evolves with each user, model-agnostic orchestration, and a context layer that decides what matters for this user in this moment. The agent stops starting from zero every conversation.

How is Sonzai different from Mem0, Letta, or Zep?

Memory-only tools retrieve facts. Agent frameworks run loops. Sonzai is the relational harness — memory, personality, mood, relationship state, knowledge, MCP access, SDKs, evals, and deployment patterns wired together as one managed platform for AI products where users come back because the agent remembers them.

Can I keep my existing model and tools?

Yes. Sonzai is model-agnostic by default. Let the Managed Runtime run the model call, pick a supported provider, bring your own provider key, or keep your own LLM loop and use Sonzai purely as the relationship and context layer. Swap the model. Keep the relationship.

Which LLM providers does Sonzai support?

Managed routing and custom model paths across OpenAI, Anthropic, Google/Gemini, xAI, OpenAI-compatible endpoints, and custom endpoints. Sonzai provides the relationship and state layer; the generating model can change as your stack changes — accumulated memory and personality persist across model swaps.

Do I have to rewire my agent to get value from Sonzai?

No. Start by sending finished turns or transcripts into Sonzai and let the relationship layer accumulate quietly. When you're ready, route assembled context back into your agent, switch to Managed Runtime, or expose the same relational state through MCP.

Why model-agnostic?

The best model today won't be the best model in six months. Your moat isn't the model — it's what your AI has learned about each user. Sonzai keeps the relationship intelligence intact as you swap models underneath.