What is the A2A (Agent2Agent) protocol?

A2A is the open protocol that lets agents built on different frameworks call each other. Agents publish an Agent Card describing their skills, and other agents read that card to decide whether and how to invoke them.

How do I make my existing agent A2A compliant?

Run super a2a adapt --entrypoint : --framework . SuperOptiX reads your agent's structure, derives its skills, and writes an A2A Agent Card plus a conformant server. Your agent code is not modified.

Which agent frameworks does SuperOptiX support for A2A?

Eight: DSPy, CrewAI, the OpenAI Agents SDK, Pydantic AI, Google ADK, the Claude Agent SDK, DeepAgents and the Microsoft Agent Framework. Adaptation works on agents written directly against those libraries, whether or not SuperOptiX built them.

How conformant is the SuperOptiX A2A implementation?

It records zero failures against the official A2A Technology Compatibility Kit. Every requirement the TCK exercises against the conformance harness passes: 73 of 73 at MUST, 7 of 7 at SHOULD, 4 of 4 at MAY. The TCK also prints a headline percentage of 77.7% at MUST, which counts 25 requirements it cannot exercise here, covering authentication, TLS, Agent Card signing and gRPC.

How do I make my agent more discoverable to other agents?

Optimize the Agent Card text. SuperOptiX scores discovery rate, invocation rate and confusion rate over a query set, then uses GEPA to rewrite skill names, descriptions and tags against that score. Agent behaviour is unchanged.

Is SuperOptiX free and open source?

Yes, under the MIT License. The A2A implementation, the adapt command, the conformance suite, the routing optimizer, the SuperSpec compiler and all eight framework integrations are in the open source package. Install with uv tool install superoptix.

Is there a live A2A agent I can call?

Yes. A SuperOptiX agent runs at https://a2a.superoptix.ai on Google Cloud Run, with its Agent Card published at https://superoptix.ai/.well-known/agent-card.json.

Who builds SuperOptiX?

Superagentic AI, which builds agentic AI infrastructure and offers support and custom engineering engagements.

CrewAI Integration

CrewAI

Multi-Agent Framework with 100,000+ Certified Developers! Role-based agent design perfect for optimization.

Works with Ollama (unlike DeepAgents!)

GEPA optimizes role + goal + backstory

100K+ Certified Developers

🔄 How It Works

SuperOptiX Workflow with CrewAI

📝SuperSpec YAML

compile

🐍CrewAI Python

evaluate

BDD Scenarios

optimize

🧬GEPA Optimizer

run

🚀crew.kickoff()

Role-Based Design

Define agents by role, goal, and backstory

Multi-Agent Native

Start single, scale to teams

Task Delegation

Assign specific tasks to agents

5 Optimizable Vars

Role + goal + backstory + task optimization

📦 Installation

uv pip install superoptix "crewai>=1.15"

Includes:

• crewai 1.15+
• SuperOptiX core with GEPA 0.1.4

Requirements:

• Python 3.11+
• Git (for DSPy dependency)

🚀 Quick Start

Pull Demo Agent

super agent pull researcher_crew

Compile

super agent compile researcher_crew --framework crewai

Evaluate

super agent evaluate researcher_crew
# Result: 4/4 PASS (100%)! 🎉

Optimize

super agent compile researcher_crew --framework crewai --optimize
super agent optimize researcher_crew --auto light --framework crewai

Run

super agent run researcher_crew --framework crewai --goal "AI trends in 2025"

🎯 What GEPA Can Optimize

GEPA optimizes role + goal + backstory together as a single variable

Component	Impact	Example
role	HIGH	"Senior AI Researcher" → "AI Research Specialist with expertise in..."
goal	HIGH	"Uncover developments" → "Discover, analyze, and synthesize breakthrough..."
backstory	MEDIUM	"You're a researcher" → "Distinguished researcher with 10+ years experience..."
task.description	HIGH	What the task should accomplish
task.expected_output	HIGH	What the task should produce

📋 SuperSpec Playbook Structure

apiVersion: agent/v1
kind: AgentSpec
metadata:
  name: researcher_crew
  id: researcher_crew
  namespace: demo
  version: 1.0.0

spec:
  target_framework: crewai
  
  # LLM Configuration - Works with Ollama!
  language_model:
    provider: ollama
    model: ollama:gpt-oss:20b
    api_base: http://localhost:11434
  
  # CrewAI Agent Profile (GEPA optimizes this!)
  persona:
    role: Senior AI Researcher
    goal: Uncover cutting-edge AI developments
    backstory: |
      You're a seasoned researcher with deep expertise in AI.
      Known for finding authoritative sources and presenting
      clear, actionable insights.
    traits:
      - analytical
      - thorough
  
  # CrewAI Task Configuration
  tasks:
    - name: research_task
      description: |
        Conduct comprehensive research on {topic}.
        Find the most relevant and up-to-date information.
      expected_output: |
        A detailed research report with:
        - 5-10 key findings
        - Clear explanations
        - Relevant context
  
  # BDD Scenarios for Testing
  feature_specifications:
    scenarios:
      - name: AI frameworks research
        input:
          topic: "AI agent frameworks"
        expected_output:
          report: "Research report"
          expected_keywords:
            - agent
            - framework
            - AI

🦙 Ollama Configuration

Why Ollama Works with CrewAI (unlike DeepAgents)

CrewAI has its own LLM class!

from crewai.llm import LLM

llm = LLM(
    model="ollama/gpt-oss:20b",
    base_url="http://localhost:11434"
)

# No bind_tools() limitations!

Recommended Ollama Models

Model	Size	Best For
gpt-oss:20b	20B	⭐ Recommended
gpt-oss:120b	120B	Maximum quality
llama3.1:8b	8B	Fast, testing
qwen2.5:14b	14B	Good balance

☁️ Cloud Model Configuration

Use cloud LLMs for production-grade performance — supports ALL LiteLLM providers

🟢

Google Gemini

FREE Tier Available

# Set API key
export GOOGLE_API_KEY="your-key"

# Playbook config:
language_model:
  location: cloud
  provider: google
  model: gemini-2.0-flash

🔵

OpenAI

# Set API key
export OPENAI_API_KEY="sk-..."

# Playbook config:
language_model:
  location: cloud
  provider: openai
  model: gpt-4o

🟠

Anthropic Claude

# Set API key
export ANTHROPIC_API_KEY="sk-ant-..."

# Playbook config:
language_model:
  location: cloud
  provider: anthropic
  model: claude-sonnet-4-5

CrewAI vs Other Frameworks

Feature	DSPy	DeepAgents	OpenAI SDK	CrewAI
Ollama Support	❌
Multi-Agent	❌	Sub-agents	Handoffs	Native
Role-Based	❌	❌	❌
Optimizable Vars	10+	1	1	5
Community	5K	1K	50K+	100K+
Best For	Prompts	Planning	Simple	Teams

Best for:

Multi-agent collaboration
Role-based task distribution
Complex crew dynamics
Team-based workflows
Production systems

❌ Not ideal for:

Simple single-agent tasks (use DSPy or OpenAI SDK)
Maximum prompt optimization (use DSPy)
Planning-heavy workflows (use DeepAgents)

📚 Additional Resources

CrewAI Documentation

Official framework docs

CrewAI GitHub

Source code

SuperOptiX Integration

Full CrewAI guide

Advanced Task Optimization

Task + Agent optimization

Ready to Build Multi-Agent Teams?

Start building collaborative AI agents with CrewAI today!