By TAKIZEN

Orchestrate AI Agents with Code

Declarative DSL for multi-agent workflows. Define complex pipelines with YAML simplicity and Python power. Native V0DN3T MCP integration.

Get Started View Examples

name: analyze_opportunity

agents:
  researcher:
    model: gpt-4o-mini
    instruction: "Search data"
    output: market_data

  analyst:
    model: claude-3-haiku
    depends_on: researcher
    instruction: "Analyze"
    output: report

on_error:
  notify: "https://..."
  fallback: researcher

Powerful Orchestration

Everything you need

Build sophisticated AI agent workflows with these powerful features

📝

Declarative YAML

Define workflows in simple, portable YAML. Perfect for version control and team collaboration.

🐍

Python Fluent API

Builder pattern with method chaining for programmatic workflow construction.

✅

Complete Validation

Detect syntactic and semantic errors before execution. Cycle detection included.

🔄

Retry Policies

Configurable retries with conditions and feedback. Handle failures gracefully.

🔌

MCP Integration

Native V0DN3T MCP Hub integration. Access web search, blockchain tools, and more.

📦

JSON IR

Both YAML and Python compile to the same Intermediate Representation.

Simple & Intuitive

Define workflows in YAML or Python

workflow.yaml

name: analyze_opportunity

agents:
  researcher:
    model: gpt-4o-mini
    tools: [web_search, fetch_url]
    instruction: "Search for data about {input.topic}"
    output: market_data

  analyst:
    model: claude-3-haiku
    depends_on: researcher
    instruction: "Analyze {market_data}"
    output: report
    retry:
      max: 2
      condition: "{decision.approved} == false"

  reviewer:
    model: gpt-4o
    depends_on: analyst
    instruction: "Approve or reject {report}"
    output: decision

on_error:
  notify: "https://hooks.takizen.io/alerts"
  fallback: researcher

workflow.py

from agentflow import WorkflowBuilder, RetryPolicy

wf = (
    WorkflowBuilder("analyze_opportunity")
    .agent("researcher",
        model="gpt-4o-mini",
        tools=["web_search", "fetch_url"],
        instruction="Search data",
        output="market_data")
    .agent("analyst",
        model="claude-3-haiku",
        depends_on="researcher",
        instruction="Analyze {market_data}",
        output="report",
        retry=RetryPolicy(
            max=2,
            condition="{decision.approved} == false"))
    .agent("reviewer",
        model="gpt-4o",
        depends_on="analyst",
        instruction="Review {report}",
        output="decision")
    .on_error(
        notify="https://hooks.takizen.io/alerts",
        fallback="researcher")
    .build()
)

V0DN3T MCP Integration

Leverage the full MCP ecosystem

AgentFlow workflows can natively access V0DN3T MCP Hub tools

web_search Tavily web search

deep_research Perplexity research

auto_fix Code fixing

podman_ps Container mgmt

d1_query Database queries

blockchain_status Web3 operations

code_interpreter E2B execution

sequential_thinking Structured reasoning