Introduction
What Tutti is, why it exists, and how it works
Tutti is an open-source multi-agent orchestration framework for TypeScript. Define a team of AI agents, give them voices (tools and integrations), and let them work together.
import { TuttiRuntime, AnthropicProvider, defineScore } from "@tuttiai/core";
import { FilesystemVoice } from "@tuttiai/filesystem";
import { GitHubVoice } from "@tuttiai/github";
const score = defineScore({
provider: new AnthropicProvider(),
agents: {
coder: {
name: "Coder",
system_prompt: "You are a senior TypeScript developer.",
voices: [new FilesystemVoice(), new GitHubVoice()],
permissions: ["filesystem", "network"],
},
},
});
const tutti = new TuttiRuntime(score);
const result = await tutti.run("coder", "Fix the bug in src/index.ts");
console.log(result.output);
Why Tutti?
LLM APIs give you text in, text out. But real tasks need agents that can read files, call APIs, control browsers, and delegate to each other. Tutti gives you:
- Agents — LLM-powered workers with system prompts and capabilities
- Voices — pluggable tool packages (filesystem, GitHub, Playwright, or build your own)
- Orchestration — an orchestrator agent can delegate tasks to specialist agents
- Observability — every LLM call, tool execution, and delegation is emitted as a typed event
- Security — permission enforcement, secret redaction, prompt injection detection, token budgets
How it works
- You write a score file (
tutti.score.ts) that defines your agents and their voices - The runtime runs the agentic loop: send messages to the LLM, execute tool calls, repeat
- Voices provide the tools that agents can call (read files, create GitHub issues, navigate browsers)
- The event bus emits typed events at every step for logging, debugging, and monitoring
What’s a “Score”?
In music, a score tells each instrument what to play. In Tutti, a score tells each agent what to do — their role, their tools, and their boundaries. The name “Tutti” means “all together” in Italian, the musical direction for the entire ensemble to play.
Next steps
- Installation — set up your environment
- Quick Start — a working agent in under 5 minutes
- Core Concepts — understand agents, voices, and scores