Features
The Relationship Layer, by capability.
Four pillars hold up every relational agent Sonzai ships — companions, characters, avatars, NPCs, brand voices. Each one is a composable layer you can mix into your stack — or take all four as the engine.
Memory & personality
Believability
Persistent memory, evolving personality, and relationship state that compound per user. Episodic memory, Big Five personality, and 4D mood — agents that don't reset on the next turn.
- Episodic memory
- Big Five personality
- 4D mood model
- Relational state
Compounding domain memory
Knowledge
A knowledge graph that grows: facts, entities, and relationships extracted from every conversation. Background learning that turns raw context into reusable memory about each user and their world.
- Knowledge graph
- Entity extraction
- Fact extraction
- Background learning
Quality at scale
Ops & Evaluation
Production observability for relational agents: evaluation rubrics, quality scorecards, and trace-level insight. Catch regressions, measure compounding gains, ship consumer-facing agents with confidence.
- Evaluation rubrics
- Quality scorecards
- Agent traces
- Regression detection
SDKs, MCP, REST
Developer Tools
TypeScript, Python, and Go SDKs, an MCP server, and a REST API for the relationship layer. Workbench for inspecting agent state, prompts, memory, and tool calls in real time.
- TS / Python / Go SDKs
- MCP server
- REST API
- Live workbench
Plug it into your stack
Six ways to wire it in.
SDKs, MCP, agent-framework plugins, an editor skill, or the raw REST API. Same relationship layer underneath — pick whichever fits the agent you're already building.
SDK
TypeScript, Python & Go
First-party SDKs with full type coverage. Same relationship-layer API across all three — pick whichever your stack runs on.
sonzai-typescript · sonzai-python · sonzai-go
Protocol
MCP
A hosted Streamable HTTP MCP endpoint. Claude Code, Cursor, Claude Desktop, ChatGPT, or any MCP-compatible client connects natively — no glue code.
api.sonz.ai/mcp/memory/{agent_id}
Plugin
OpenClaw
Drop the sonzai-openclaw plugin into an OpenClaw project and Sonzai becomes the agent's contextEngine — persistent memory, mood, personality, relationships, all under OpenClaw's chat loop.
sonzai-openclaw
Plugin
Hermes
Drop the sonzai-hermes plugin into a Hermes agent project. Sonzai owns the relationship layer; Hermes keeps owning orchestration and tools.
sonzai-hermes
Skill
Claude Code Skill
A drift-checked skill for Claude Code, Codex, and Gemini CLI that writes Sonzai SDK calls correctly — validated against the live api.sonz.ai OpenAPI spec.
github.com/sonz-ai/sonzai-claude-skill
API
REST
Every SDK is a thin wrapper. The raw REST API is documented and stable — useful when you're on a stack the SDKs don't cover yet.
OpenAPI 3.1 · 51 tags
Deeper, by feature
Read how each piece works.
Memory
Episodic, semantic, and relational memory.
Personality
Big Five traits that drift with experience.
Emotions
Four-dimensional mood model.
Self-improvement
Agents that learn from their own traces.
Multiplayer memory
Shared memory across agents and users.
Wake-ups
Proactive triggers without a request.
Voice
Real-time voice with mood-aware delivery.
Scheduled reminders
Time-aware action and follow-ups.
See the full reference.
Every feature is documented with examples in TypeScript, Python, Go, and via MCP. Quickstarts ship in under five minutes.