behavioralfingerprint 0.6.0

Capture how your AI agent behaves at deployment. Monitor how that behavior changes over time.

behavioralfingerprint

Capture how your AI agent behaves at deployment. Monitor how that behavior changes over time.

Part of the Thread Suite — AI agent reliability infrastructure.

Install

pip install behavioralfingerprint

Quick Start

from behavioralfingerprint import BehavioralFingerprint

bf = BehavioralFingerprint()  # defaults to https://behavioral-fingerprint.onrender.com

# Register your agent
bf.register_agent(
    agent_id="my-agent-v1",
    endpoint_url="https://your-agent.com/run",
    name="My Production Agent"
)

# Capture a behavioral fingerprint
result = bf.capture_fingerprint("my-agent-v1")
print(result["scores"])
# {
#   "verbosity_score": 0.42,
#   "hedging_rate": 0.18,
#   "refusal_rate": 0.0,
#   "confidence_score": 0.71,
#   "consistency_score": 0.63,
#   "adherence_score": 0.90
# }

# Get latest fingerprint
latest = bf.get_latest_fingerprint("my-agent-v1")

# View fingerprint history
history = bf.list_fingerprints("my-agent-v1")

Agent Endpoint Contract

Your agent endpoint must accept POST requests with:

{"input": "probe text here"}

And return JSON with an output, response, result, or content field:

{"output": "the agent's response"}

Six Behavioral Dimensions

Dimension	What it measures
verbosity_score	How much the agent says (0 = concise, 1 = verbose)
hedging_rate	How often the agent qualifies its answers
refusal_rate	How often the agent declines requests
confidence_score	Ratio of confident to uncertain language
consistency_score	Similarity across semantically equivalent inputs
adherence_score	Compliance with explicit formatting/length instructions

Thread Suite

Iron-Thread · TestThread · PromptThread · ChainThread · PolicyThread · ThreadWatch · Behavioral Fingerprint

"Built for the age of AI agents."