Sunesis Agents
Agents that understand, not just respond
Build and deploy conversational agents with persistent memory, function calling, and RAG. Sub-500ms turn-taking that feels like a real conversation, not a chatbot.
Architecture
The agent pipeline
Every turn flows through four stages, listen, think, act, speak, in under 500ms.
Speech InSTT + emotion
ReasonMemory + RAG
ActFunction calls
SpeakTTS · <200ms
End-to-end median: 412ms · p99: 680ms
Capabilities
Memory, tools, and grounding
Persistent Memory
Long-term memory across sessions. Agents recall preferences, history, and context, they never start from scratch.
Function Calling
Connect any API, database, or service. Agents call tools autonomously to get real work done.
RAG
Ground every response in your own knowledge base. Retrieval-augmented generation keeps answers accurate.
Sub-500ms Turn-Taking
Barge-in, interruption handling, and real-time turn-taking that feels like talking to a person.
Developer-First
Define an agent in 10 lines
agent.ts
import { Sunesis } from "@sunesis/agents";
const agent = new Sunesis.Agent({
name: "Aria",
voice: "clone:voice_abc123",
memory: { type: "long-term" },
tools: [sendEmail, bookMeeting, searchKB],
});
agent.on("message", async (ctx) => {
const reply = await agent.reason(ctx); // memory + RAG
await agent.speak(reply); // <200ms TTS
});