§INTEGRATION · CURSOR
Cursor, agent mode, without the gap.
Cursor's settings let you override the OpenAI / Anthropic base URL under the model configuration. Point it at Sentisec and proposed agent actions flow through our control plane first. Best-effort compatibility — Cursor changes quickly and we track it.
ONE-LINE SETUP
Two lines. One terminal. Done.
No shell-rc change needed for Cursor; the override is a per-app setting. The installer is optional but helps diagnose connectivity.
bash
$ curl -fsSL https://get.sentisec.ch | bash $ Cursor → Settings → Models → override base URL to: https://api.sentisec.ch/v1
PATH OF A TOOL CALL
One hop. No framework rewrite.
Diagram is intentionally kept at the category level — the detail of the checks lives on the developer landing page and under NDA.
cursor (agent mode) │ ▼ proposed actions sentisec control plane │ cognitive integrity score · workspace policy ▼ model vendor API │ ▼ tool result back to cursor — or halt + reason
§CATCHES
What this catches
- Cursor agent-mode Bash / shell actions that cross the workspace boundary.
- Web-browsing actions whose arguments contain URLs outside your workspace policy.
- Edit / Read operations against sensitive path patterns (`.env`, `.ssh/`, `credentials*`).
§LIMITS
What this doesn't (yet)
- Cursor changes quickly. We commit to regular re-verification; between releases, new action shapes may briefly be best-effort.
- If a Cursor feature executes entirely on the vendor side, an external control plane cannot see it — the Codex / OpenAI Responses caveat applies.