HumanMirror Agent Control Plane for Dify: independent control before consequential agent actions, with deterministic decisions and signed governance receipts.
HumanMirror Guard is the Dify entry point to the HumanMirror Agent Control Plane: an independent pre-execution checkpoint for workflows that can trigger consequential actions.
The anonymous demo stores no raw action text in the HumanMirror application database and does not expose payment policy, authorization evidence, state continuity, perception, custom policy terms, or enterprise context.
Send the intended action immediately before a tool or workflow side effect. HumanMirror returns a deterministic governance result:
The response also includes risk details and a signed receipt that can be retained with workflow evidence.
The API-key field is optional for the first bounded demo calls.
Start self-serve access — 14-day trial, no card required:
https://humanmirror.fr/developers/
Enterprise Guard:
https://humanmirror.fr/enterprise-guard/
The plugin requires outbound HTTPS access to the fixed HumanMirror Enterprise Guard API at . It does not access arbitrary URLs, local files, a shell, or a database.
Send only the context required to evaluate the intended action. Do not place credentials, passwords, private keys, authentication cookies, or other secrets in .
See for the plugin-specific data-handling disclosure.
https://github.com/VibeMirror-coder/humanmirror/tree/main/integrations/dify-humanmirror
HumanMirror: https://humanmirror.fr/
Contact: [email protected]
Self-serve Developer keys expose the bounded Dify action-preflight surface only. HumanMirror Fleet is designed for recurring multi-agent workloads; payment policy, authorization evidence, perception, state continuity and industry/jurisdiction context remain Enterprise Control Plane / Enterprise Guard capabilities.
When no API key is configured, HumanMirror Guard calls the fixed demo endpoint at . The demo accepts only an agent identifier and bounded action text. It does not accept financial policy, payment details, authorization evidence, perception, state hashes, enterprise headers, or custom required/forbidden terms. The response is marked and includes the remaining anonymous quota plus the Developer upgrade path.
Some outbound calls build their address at runtime, so the list may be incomplete.
Last checked 10/14/2026 06:03 PM · version 0.1.3