Drop-in replacement

Sentisec is a drop-in replacement for the Anthropic Messages API and the OpenAI Chat Completions API. You change the base URL your client already has and swap in a Sentisec key. There is no SDK to install, no import to add, and no code path in your application that knows Sentisec exists.

The swap

from anthropic import Anthropic

# Before:
client = Anthropic(api_key="sk-ant-...")

# After:
client = Anthropic(
    base_url="https://api.sentisec.ch/ws/<your-workspace-id>",
    api_key="sk_sentisec_xxxxxxxx_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
)

from openai import OpenAI

# Before:
client = OpenAI(api_key="sk-...")

# After:
client = OpenAI(
    base_url="https://api.sentisec.ch/ws/<your-workspace-id>/v1",
    api_key="sk_sentisec_xxxxxxxx_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",
)

export ANTHROPIC_BASE_URL=https://api.sentisec.ch/ws/<your-workspace-id>
export ANTHROPIC_API_KEY=sk_sentisec_xxxxxxxx_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
claude

# Settings → Models → Custom OpenAI
Base URL:  https://api.sentisec.ch/ws/<your-workspace-id>/v1
API Key:   sk_sentisec_xxxxxxxx_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Note the /v1 suffix: it belongs on the OpenAI-shaped URL and not on the Anthropic-shaped one, because each client appends a different path.

What stays exactly the same

• The model id you send.
• The request body, field for field.
• The response shape — Sentisec preserves the upstream's format byte-for-byte unless it intervenes.
• Streaming: "stream": true on the same endpoint, with frame boundaries preserved.
• Tool-call schemas and vision content parts.

What Sentisec adds

• A verdict on each step, visible in the dashboard.
• Forensic evidence for flagged activity.
• A tool call that would exfiltrate data or take a destructive action is stopped before it runs, and the answer explains why. The CLI agents overview shows exactly what that looks like on the wire.
• Per-workspace controls over what gets enforced.

Which upstreams you can point at

Six providers carry traffic through this drop-in today:

The URL is the same for every OpenAI-shaped provider — Sentisec picks the upstream from the model id, or from the X-Sentisec-Upstream-Provider header when you want to pin it explicitly. Vertex AI always needs pinning, because its model ids overlap other vendors' naming. See providers for the per-provider setup.

Azure OpenAI is not available yet. If it is a requirement for your deployment, tell us.

Running your own Sentisec plane? Substitute your own host for https://api.sentisec.ch and keep the rest of the path — see enterprise access.

Related

• Providers — per-upstream setup
• Integrations — the same swap from inside LangChain, LlamaIndex, CrewAI, and friends
• Quickstart — first monitored call end to end