quint-agent

Behavioral telemetry and policy enforcement for AI agents.

Capture every prompt, tool call, and tool result from your AI agents. Enforce YAML-defined policy at the tool-call boundary. Stream telemetry to the Quint cloud for behavioral analysis and divergence detection.

Install

pip install quint-agent

With auto-instrumentation (zero code change):

pip install 'quint-agent[otel-openai]'     # auto-capture OpenAI calls
pip install 'quint-agent[otel-anthropic]'   # auto-capture Anthropic calls
pip install 'quint-agent[otel-langchain]'   # auto-capture LangChain chains
pip install 'quint-agent[otel-all]'         # all frameworks

With framework integrations:

pip install 'quint-agent[anthropic]'   # guarded_dispatch for Anthropic
pip install 'quint-agent[langchain]'   # QuintCallbackHandler
pip install 'quint-agent[mcp]'         # QuintMCPClient
pip install 'quint-agent[all]'         # everything

Zero-code observability (2 lines)

import quint_agent

quint_agent.init(api_key="qak_...")
# That's it. Every OpenAI/Anthropic/LangChain call is auto-captured.

If OTel instrumentors are installed ([otel-openai], [otel-anthropic], etc.), quint.init() auto-detects and instruments them. No decorators, no callbacks, no code changes.

@quint.guard — policy enforcement

import quint_agent

quint_agent.init(api_key="qak_...")

@quint_agent.guard()
def read_file(path: str) -> str:
    return open(path).read()

@quint_agent.guard()
def run_command(cmd: str) -> str:
    import subprocess
    return subprocess.check_output(cmd, shell=True, text=True)

read_file(path="/etc/hosts")        # allowed — captured to cloud
read_file(path="/home/me/.env")     # blocked — ToolBlockedError

Anthropic integration

import anthropic
import quint_agent
from quint_agent.integrations.anthropic import guarded_dispatch

quint_agent.init(api_key="qak_...")

tools_map = {"read_file": read_file}
client = anthropic.Anthropic()
response = client.messages.create(
    model="claude-sonnet-4-6",
    max_tokens=1024,
    tools=[{"name": "read_file", "description": "Read a file", "input_schema": {
        "type": "object", "properties": {"path": {"type": "string"}}, "required": ["path"]
    }}],
    messages=[{"role": "user", "content": "Read /etc/passwd"}],
)

for block in response.content:
    if block.type == "tool_use":
        result = guarded_dispatch(block.name, block.input, tools_map)

LangChain

from quint_agent.integrations.langchain import QuintCallbackHandler

chain.invoke({"input": "..."}, config={"callbacks": [QuintCallbackHandler()]})

MCP

from quint_agent.integrations.mcp import QuintMCPClient

async with mcp_client() as session:
    guarded = QuintMCPClient(session)
    result = await guarded.call_tool("read_file", {"path": "/x"})

Custom policy

# my_policy.yaml
rules:
  - name: no_prod_db
    action: block
    reason: "Production database access not allowed"
    when:
      - op: equals
        path: tool_name
        value: execute_sql
      - op: contains
        path: args.connection
        value: prod

quint_agent.init(api_key="qak_...", policy="my_policy.yaml")

Default policy

Ships with 4 rules out of the box:

• Block .env file reads
• Block AWS credentials access
• Escalate on secrets/ directory reads
• Block destructive shell commands (rm -rf /)

Override by passing your own policy to quint.init(policy=...).

Docs

Full documentation at docs.quintai.dev/sdk.

License

Apache 2.0 — see LICENSE.

Built by Quint Security.