Examples
Complete working recipes: Anthropic, OpenAI, Claude Code, LangChain, and n8n.
Each recipe here is end to end: the code you paste, what comes back, and where the result shows up. Pick the one closest to what you are building.
Hello, monitored OpenAI
The OpenAI Chat Completions hello-world, routed through Sentisec. The fastest first session.
Hello, monitored Anthropic
The Anthropic Messages hello-world, routed through Sentisec.
Monitor a Claude Code session
Point Claude Code at Sentisec and every bash and file-edit call earns a verdict.
Wrap a LangChain agent
A LangChain agent with the Sentisec base URL — no agent-side code changes.
n8n Gmail triage
Guard an inbox-triage automation, the textbook indirect-injection target.
What every recipe assumes
- A Sentisec workspace, and an API key from Settings → API keys.
- Your provider key stored under Settings → Providers, in Provider keys.
- Python 3.11+ for the Python samples, Node 20+ for the TypeScript ones.
Every recipe uses the workspace-scoped base URL — see your base URL for what the workspace id in the path does.
What "monitored" looks like
A recipe that makes a plain text completion produces a session with no action to adjudicate — the calm case. A recipe whose agent calls tools produces a verdict per action, and a stopped call is withheld, with an explanation in place of the tool call the model asked for. The cURL reference has the exact shapes.