agentbill-py-crewai 10.5.0

AgentBill integration for CrewAI - Zero-config crew tracking

AgentBill CrewAI Integration

OpenTelemetry-based crew tracker for automatically tracking and billing CrewAI agent usage.

Installation

pip install agentbill-crewai

This will automatically install the required dependencies:

• crewai
• agentbill-langchain (CrewAI uses LangChain callbacks under the hood)

Quick Start

from agentbill_crewai import track_crew
from crewai import Agent, Task, Crew
from langchain_openai import ChatOpenAI

# 1. Initialize LLM
llm = ChatOpenAI(model="gpt-4o-mini")

# 2. Create agents
researcher = Agent(
    role="Research Analyst",
    goal="Find and analyze data",
    backstory="Expert researcher with attention to detail",
    llm=llm
)

writer = Agent(
    role="Content Writer",
    goal="Write engaging content",
    backstory="Creative writer with storytelling skills",
    llm=llm
)

# 3. Create tasks
research_task = Task(
    description="Research the topic: {topic}",
    agent=researcher,
    expected_output="Comprehensive research findings"
)

writing_task = Task(
    description="Write an article based on the research",
    agent=writer,
    expected_output="Well-written article"
)

# 4. Create crew
crew = Crew(
    agents=[researcher, writer],
    tasks=[research_task, writing_task]
)

# 5. Run with AgentBill tracking!
result = track_crew(
    crew=crew,
    inputs={"topic": "AI in healthcare"},
    agentbill_config={
        "api_key": "agb_your_api_key_here",
        "base_url": "https://api.agentbill.io",
        "customer_id": "customer-123",
        "debug": True
    }
)

print(result)

# ✅ Automatically captured:
# - All agent LLM calls
# - Token usage per agent
# - Task execution times
# - Total crew cost
# - Agent-level profitability

Features

• ✅ Zero-config instrumentation - Just wrap with track_crew()
• ✅ Agent-level tracking - Track each agent's LLM usage
• ✅ Task-level metrics - Measure task execution time
• ✅ Multi-agent support - Track complex multi-agent workflows
• ✅ Cost calculation - Auto-calculates costs per agent
• ✅ Crew profitability - Compare crew costs vs revenue
• ✅ OpenTelemetry compatible - Standard observability

Advanced Usage

Track Revenue Per Crew

result = track_crew(
    crew=crew,
    inputs={"topic": "AI trends"},
    agentbill_config={
        "api_key": "agb_...",
        "base_url": "https://...",
        "customer_id": "customer-123"
    },
    revenue=5.00,  # What you charged for this crew execution
    revenue_metadata={
        "subscription": "enterprise",
        "feature": "research_crew"
    }
)

Use with Custom LLMs

from langchain_anthropic import ChatAnthropic

# Works with any LangChain-compatible LLM
anthropic_llm = ChatAnthropic(model="claude-3-5-sonnet-20241022")

agent = Agent(
    role="Analyst",
    goal="Analyze data",
    backstory="Expert analyst",
    llm=anthropic_llm  # CrewAI auto-tracks this!
)

Sequential vs Parallel Crews

# Sequential crew (default)
sequential_crew = Crew(
    agents=[agent1, agent2],
    tasks=[task1, task2],
    process=Process.sequential  # Tasks run one after another
)

# Parallel crew
from crewai import Process

parallel_crew = Crew(
    agents=[agent1, agent2],
    tasks=[task1, task2],
    process=Process.parallel  # Tasks run concurrently
)

# Both tracked automatically!
track_crew(sequential_crew, {...})
track_crew(parallel_crew, {...})

Hierarchical Crews

# Manager agent delegates to worker agents
manager = Agent(
    role="Project Manager",
    goal="Coordinate the team",
    backstory="Experienced manager",
    llm=llm
)

crew = Crew(
    agents=[manager, worker1, worker2],
    tasks=[task1, task2],
    process=Process.hierarchical,  # Manager delegates
    manager_llm=llm
)

# All agent interactions tracked!
result = track_crew(crew, inputs={...}, agentbill_config={...})

Configuration

agentbill_config = {
    "api_key": "agb_...",           # Required - get from dashboard
    "base_url": "https://...",      # Required - your AgentBill instance
    "customer_id": "customer-123",  # Optional - for multi-tenant apps
    "account_id": "account-456",    # Optional - for account-level tracking
    "debug": True,                  # Optional - enable debug logging
    "batch_size": 10,               # Optional - batch signals before sending
    "flush_interval": 5.0           # Optional - flush interval in seconds
}

How It Works

The crew tracker wraps CrewAI execution:

1. Inject Callback - Adds AgentBill callback to all agents' LLMs
2. Track Agents - Monitors each agent's LLM calls
3. Track Tasks - Measures task execution time
4. Calculate Costs - Sums up all agent costs
5. Send Signals - Sends data to AgentBill via the unified OTEL pipeline

All agent interactions are automatically captured without code changes.

Real-World Example: Research Crew

from agentbill_crewai import track_crew
from crewai import Agent, Task, Crew
from crewai_tools import SerperDevTool
from langchain_openai import ChatOpenAI

# Tools
search_tool = SerperDevTool()
llm = ChatOpenAI(model="gpt-4o-mini")

# Agents
researcher = Agent(
    role="Senior Research Analyst",
    goal="Discover cutting-edge developments in {topic}",
    backstory="Veteran researcher with 10+ years experience",
    tools=[search_tool],
    llm=llm
)

analyst = Agent(
    role="Data Analyst",
    goal="Analyze research findings and extract insights",
    backstory="Expert at data analysis and pattern recognition",
    llm=llm
)

writer = Agent(
    role="Content Writer",
    goal="Create compelling content from insights",
    backstory="Award-winning writer with storytelling expertise",
    llm=llm
)

# Tasks
research_task = Task(
    description="Research {topic} and compile findings",
    agent=researcher,
    expected_output="Comprehensive research report"
)

analysis_task = Task(
    description="Analyze research and identify key insights",
    agent=analyst,
    expected_output="Detailed analysis with insights"
)

writing_task = Task(
    description="Write engaging article from analysis",
    agent=writer,
    expected_output="Publication-ready article"
)

# Crew
research_crew = Crew(
    agents=[researcher, analyst, writer],
    tasks=[research_task, analysis_task, writing_task],
    verbose=True
)

# Execute with tracking
result = track_crew(
    crew=research_crew,
    inputs={"topic": "Quantum Computing in Drug Discovery"},
    agentbill_config={
        "api_key": "agb_your_key",
        "base_url": "https://api.agentbill.io",
        "customer_id": "pharma-corp-123"
    },
    revenue=50.00,  # What you charged for this research
    revenue_metadata={
        "client": "PharmaCorp",
        "project": "drug_discovery_research"
    }
)

print("Article:", result)

# ✅ Dashboard shows:
# - Cost per agent (researcher, analyst, writer)
# - Total crew cost
# - Revenue ($50)
# - Net margin (revenue - cost)
# - Agent efficiency metrics

Troubleshooting

Not seeing agent data?

1. Ensure CrewAI agents have LLMs assigned
2. Check API key is correct
3. Enable debug: True to see logs
4. Verify crew is actually running (not just created)

Missing token counts?

• Some LLMs don't return usage data
• OpenAI and Anthropic provide accurate counts
• Local models may need manual instrumentation

Multiple crews running?

Each track_crew() call is independent - perfect for parallel crew execution!

Idempotent Ingestion (X-Request-Id)

Every OTLP batch this SDK exports to /functions/v1/otel-collector carries an X-Request-Id: <uuid4> header (auto-generated per export, reused across transport-level retries within that export). The collector deduplicates on (company_id, request_id), so a transient 5xx / network retry will not double-write WAL rows, spans, or billing events.

• No configuration required — the header is emitted automatically by the v7.x line of this SDK.
• Server fallback: clients that cannot set custom headers may send {"requestId": "..."} at the JSON envelope root.
• Native OTLP exporters: the standard OTLP-HTTP exporter keeps headers static across batches, so per-batch rotation is not yet wired on that path (PHASE3-R3-SDK-REQUESTID-NATIVE — tracked); the collector still dedupes those payloads on the natural key (company_id, trace_id, span_id).

See the OpenTelemetry Integration docs for the full request-id contract.

License

MIT