bzzz-cli 0.1.0

Bzzz CLI - Command line interface for Agent orchestration
bzzz-cli-0.1.0 is not a library.

Bzzz

License: MIT Tests Rust GitHub release

The trusted middleware for Agent capabilities.

Define → Orchestrate → Execute

From single agents to complex swarms, across any runtime.


What is Bzzz?

Bzzz is not another developer-first multi-agent workflow framework. It is:

  • Protocol-first: AgentSpec, SwarmFile, RuntimeAdapter define clear contracts
  • Runtime-agnostic: Execute on Local, Docker, HTTP, or custom runtimes
  • Pattern-based: 9 orchestration patterns for any scenario
  • Agent-ready: Designed to be called by your primary Agent

Ecosystem Position

┌─────────────────────────────────────────┐
│  Agent Products                         │  Claude Code, Codex, Cursor
├─────────────────────────────────────────┤
│  Agent Orchestration ← Bzzz 在这里      │  SwarmFile + 编排引擎
├─────────────────────────────────────────┤
│  Agent Communication                    │  A2A 协议(Agent 发现 + 通信)
├─────────────────────────────────────────┤
│  Agent-Tool Connection                  │  MCP(Agent 用工具)
├─────────────────────────────────────────┤
│  Agent Frameworks                       │  LangGraph, CrewAI, Claude SDK
├─────────────────────────────────────────┤
│  LLM APIs                              │  Claude API, OpenAI API
└─────────────────────────────────────────┘

类比

  • MCP: Agent 怎么调用工具
  • A2A: Agent 怎么被发现、怎么互相通信
  • Bzzz: 多个 Agent 怎么按工作流协作 ← A2A 生态里缺的这一块

Quick Start

# Clone
git clone https://github.com/fengjunhui/bzzz.git
cd bzzz

# Build
cargo build --release

# Run an example
cargo run -- run -f examples/basics/sequence.swarm.yaml

Demo

CLI Demo Recording

View the CLI demo using asciinema:

asciinema play docs/demos/bzzz-demo.cast

Or view the plain text output:

$ bzzz run --help

Run an Agent or Swarm

Usage: bzzz run [OPTIONS] --file <FILE>

Options:
  -f, --file <FILE>      Path to Agent Spec or SwarmFile
  -b, --background       Run in background
  -t, --timeout <TIMEOUT> Timeout in seconds
  -r, --runtime <RUNTIME> Runtime to use (local, docker, http)
  ...

$ bzzz run -f examples/basics/sequence.swarm.yaml

🚀 Running: examples/basics/sequence.swarm.yaml
  Swarm: simple-sequence (simple task)
  Worker: worker-a
  Runtime: Local
  Status: Completed
  Output: {"stdout": "done"}

Core Concepts

Concept Purpose
AgentSpec Declare a single agent's capability
SwarmFile Orchestrate multiple agents with patterns
RuntimeAdapter Execute across different runtimes
Pattern 9 modes: sequence, parallel, conditional, loop, delegate, supervisor, compete, escalation, alongside

Example

# agent.yaml
apiVersion: bzzz.dev/v1
kind: agent
id: my-agent
runtime:
  kind: Local
  config:
    command: "echo 'Hello from Bzzz!'"
# swarm.yaml
apiVersion: bzzz.dev/v1
kind: swarm
id: my-swarm
workers:
  - name: my-agent
    spec: agent.yaml
flow:
  type: sequence
  steps:
    - my-agent
# Execute
bzzz run -f swarm.yaml

CLI Commands

Command Description
bzzz run -f <file> Run AgentSpec or SwarmFile
bzzz run --background Run in background
bzzz status --id <id> Query execution status
bzzz stop --id <id> Stop execution
bzzz list List all executions
bzzz validate -f <file> Validate configuration

Documentation

📖 Full Documentation: docs-site/content/ (mdBook format)

Quick Links

Specification Docs

Status

Phase Status
Phase 0: Data Flow ✅ Complete
Phase 1: Protocol Freeze ✅ Complete
Phase 2: First Demo ✅ Complete
Phase 3: CLI Context ✅ Complete
Phase 4: Cross-Framework ✅ Complete
Phase 5: Developer Experience ✅ Complete
Phase 6: Open Source Release ✅ Complete

License

Licensed under MIT License.

Contributing

See CONTRIBUTING.md for guidelines.


Bzzz - Turning compute capabilities into Agent capabilities.