Author: tomphillipslabs Type: tool Source repository: https://github.com/TomPhillipsLabs/dify-sentinelproxy
Self-healing evaluation for AI agents. Mirror each agent turn to SentinelProxy — its judge catches hallucinations, prompt injections, PII leaks and competitor mentions, then hands you a hardened system prompt.
- In SentinelProxy, open your project's Connect tab and copy the API key.
- In Dify, add the SentinelProxy tool and create a credential:
- Project API Key — paste the key. Click Test to confirm it works.
- Base URL — leave as
https://sentinelproxy.tomphillips.ukunless you self-host.
Add the Log Agent Turn tool node after your LLM / Agent node and map:
| Field | Map to |
|---|---|
| User Input | the incoming user message |
| LLM Response | the model's output |
| System Prompt | your agent's system prompt (optional) |
| Latency (ms) | the call latency, if available (optional) |
The node POSTs to /api/v1/ingest, which returns immediately and evaluates the
turn asynchronously — it never blocks your workflow. PII is scrubbed at the edge
before anything is stored.
See PRIVACY.md.