Quick start — inline agents

QuickAgent and the antcrew quick CLI command let you define and run a multi-agent pipeline with zero Python code. Role and goal are specified as strings — no class definition, typed artifacts, or EventBus knowledge required.

This is the fastest path from idea to running pipeline: role and goal as strings, typed artifact contracts and TraceLog underneath.

CLI

antcrew quick "Research AI agent frameworks" \
  "Researcher: Find and compare the top Python AI agent frameworks released in 2025" \
  "Analyst: Identify the 3 most significant differences from a production perspective" \
  "Writer: Write a clear 300-word summary for a technical decision-maker"

Each argument after the goal is an agent spec: "Role: description of what this agent does". Agents run sequentially; each sees the previous agent's output.

Options

Flag	Default	Description
--model	claude	LLM to use (same as antcrew run)
--json	false	Print full state JSON instead of the result panel
--push URL	—	Dispatch to a remote platform instance
--api-key KEY	$ANTCREW_API_KEY	Platform API key (required with --push)

Push to platform

antcrew quick "Build a login system" \
  "Architect: Design the auth architecture" \
  "Developer: Write the implementation plan" \
  --push https://antcrew.org \
  --api-key acw_...

Python API

from antcrew.agents.quick_agent import QuickAgent, QuickTeam
from antcrew.models.anthropic_model import AnthropicModel

llm = AnthropicModel()

# Single agent
agent = QuickAgent(llm, role="Researcher", goal="Find recent papers on RAG")
state = agent.run({"request": "What's new in vector retrieval?"})
print(state["result"])

# Multi-agent team
team = QuickTeam(
    specs=[
        "Researcher: Find and summarize recent papers on RAG",
        "Writer: Synthesize findings into a clear report",
    ],
    llm=llm,
)
result = team.run("What's new in vector retrieval?")
print(result["result"])

How it compares to BaseAgent

Feature	QuickAgent	BaseAgent subclass
Setup time	Seconds (one string)	Minutes (class + imports)
Typed artifacts	No (dict passthrough)	Yes (Pydantic contracts)
Governance hash	Yes	Yes
HITL	Not built-in	Built-in (approval_required)
Memory	Yes (_mem_get/_mem_set)	Yes
Production recommended	Prototyping	Yes

Migrating to typed agents

When a QuickAgent proves valuable and needs typed outputs or HITL:

# Start with QuickAgent:
specs = ["Researcher: Find papers on RAG"]

# Graduate to typed agent:
from antcrew.core.agent import BaseAgent
from antcrew.core.artifacts import ArtifactContract, ResearchDocument

research_contract = ArtifactContract("research_document", ResearchDocument)

class ResearchAgent(BaseAgent):
    name = "researcher"
    role_description = "Find papers on RAG"
    consumes = ["request"]
    produces = ["research_document"]

    def run(self, state):
        result = self.system("Find papers on RAG.", state["request"])
        doc = ResearchDocument(title="RAG Research", key_findings=[result])
        return research_contract.inject(doc)