Documentation Index
Fetch the complete documentation index at: https://docs.vane.build/llms.txt
Use this file to discover all available pages before exploring further.
Installation
No external dependencies. Uses only the Python standard library (urllib, threading, json).
Requires Python 3.9+.
Three-line setup
from counsel_crewai import VaneObserver
observer = VaneObserver(
base_url="https://vane.build",
api_key="counsel_a1b2c3d4...",
agent_id="contract-reviewer",
)
crew = observer.wrap(crew)
VaneObserver
Constructor
VaneObserver(
base_url: str,
api_key: str,
agent_id: str,
company_id: str = "",
)
| Parameter | Type | Required | Description |
|---|
base_url | str | Yes | Base URL of your Vane instance. Trailing slash is stripped. |
api_key | str | Yes | Company-scoped API key. |
agent_id | str | Yes | Agent identifier that appears on every record. |
company_id | str | No | Company ID embedded in attestation payloads. If empty, the API key’s company is used. |
wrap(crew)
Attaches attestation callbacks to the crew and returns it. Preserves any existing step_callback and task_callback — both the original callbacks and the attestation callbacks will be called.
crew = observer.wrap(crew)
# crew.step_callback and crew.task_callback are now wrapped
What gets attested
| Event | actionType | Payload fields |
|---|
| Tool invocation (AgentAction) | "tool-call" | tool, toolInput, observation |
| Agent finished (AgentFinish) | "agent-finish" | output (the return_values dict) |
| Any other step output | "agent-step" | raw (string representation) |
| Task completed | "task-complete" | description, agent, output |
Error handling
All HTTP calls happen in daemon threads. If the Vane server is unavailable, the error is printed to stdout and the crew execution continues uninterrupted. Attestation failures never block or crash your crew.
Full example
from crewai import Agent, Task, Crew
from crewai_tools import SerperDevTool
from counsel_crewai import VaneObserver
# Set up the observer
observer = VaneObserver(
base_url="https://vane.build",
api_key="counsel_a1b2c3d4...",
agent_id="legal-researcher",
company_id="acme",
)
# Create your crew normally
researcher = Agent(
role="Legal Researcher",
goal="Find relevant legal precedents",
backstory="Expert in EU AI regulation",
tools=[SerperDevTool()],
verbose=True,
)
task = Task(
description="Research EU AI Act compliance requirements for financial services",
agent=researcher,
expected_output="A structured summary of compliance requirements",
)
crew = Crew(agents=[researcher], tasks=[task])
# Wrap the crew — three lines total
crew = observer.wrap(crew)
result = crew.kickoff()
print(result)
Using with existing callbacks
If your crew already has step or task callbacks, they are preserved:
def my_existing_step_callback(step_output):
print(f"Step: {step_output}")
crew = Crew(
agents=[...],
tasks=[...],
step_callback=my_existing_step_callback,
)
# Both my_existing_step_callback AND counsel attestation will run
crew = observer.wrap(crew)
