Python

Sentisec speaks the Anthropic and OpenAI wire formats, so the Python integration is a base-URL change on the client you already use. No Sentisec import required.

You need: a Sentisec API key. Mint one in the dashboard under Settings → API keys, or follow the authentication guide.

Route your client through Sentisec

import os
import anthropic

client = anthropic.Anthropic(
    base_url="https://api.sentisec.ch/ws/<your-workspace-id>",
    api_key=os.environ["SENTISEC_API_KEY"],
)

resp = client.messages.create(
    model="claude-sonnet-4",
    max_tokens=256,
    messages=[{"role": "user", "content": "hello"}],
)
print(resp.content[0].text)

Note the /v1 on the OpenAI URL and not on the Anthropic one — the Anthropic SDK appends its own. Keep the workspace id in the path; see your base URL.

Every call from client now flows through Sentisec, is scored by the integrity engine, and appears as a session in your dashboard. Your upstream provider key stays in your workspace, so it never has to be present in the process making the call.

The same change works for any library that lets you set a base URL on its underlying client — LangChain, LangGraph, PydanticAI, LlamaIndex, CrewAI. See integrations for the per-framework form.

Handling a blocked tool call

Sentisec does not raise an HTTP error when it stops a tool call. The request returns 200. By default the call is withheld: it is removed from the response, and a text block in its place explains why, streamed or not.

for block in resp.content:
    if block.type == "text" and block.text.startswith("[Sentisec blocked"):
        ...  # a call was stopped; the text says why. Do not retry it.

If you use an agent framework, you need to write nothing: with no call left to run, the loop ends and the explanation is its answer. To let the model re-plan instead, a workspace admin can opt in to rename; the call then comes back as sentisec_tool_call_blocked. Full detail in what happens when a tool call is blocked.

Install the CLI

sentisec-sdk on PyPI also ships the sentisec command, which pairs this machine with your workspace and drives subscription mode.

pipx install sentisec-sdk
sentisec --version

pipx isolates the package in its own virtual environment and puts sentisec on your PATH. Full walkthrough in the CLI quickstart.

To use the package as a library instead, install it into a virtual environment:

python3 -m venv .venv
source .venv/bin/activate
pip install sentisec-sdk

The venv keeps its dependencies isolated and sidesteps the PEP 668 externally-managed-environment error on macOS, Debian, Ubuntu, and Fedora. Python 3.11 through 3.14 are supported.

Verify it worked

Open the dashboard at your workspace → Sessions. The session row is there; the Live view shows sessions as they arrive.

Verdicts attach to tool actions — the moments an agent reaches out and touches something. A plain completion like the one above makes no tool call, so it lands as a completed session with nothing to adjudicate. Once your agent starts calling tools, each action is scored:

Click any row for per-step detail.

Next