by asqav · v0.0.6
AI agent governance for Dify. Cryptographically sign agent actions into an audit trail, verify signatures, and require multi-party authorization for high-risk actions. Powered by Asqav.
This community listing does not yet include every recommended support, privacy, pricing, and permission disclosure. Review the available package permissions before installing.
Available inside your emploidai workspace after installation.
Available inside your emploidai workspace after installation.
Asqav adds AI agent governance to your Dify workflows. It checks every agent action against your policies before it runs, blocks anything out of bounds, and cryptographically signs the rest into a tamper-evident audit trail you can verify later.
Asqav is the evidence layer for AI agents. This plugin connects your Dify deployment to the Asqav API so each action an agent takes is reviewed before your workflow commits to it. An allowed action is signed with the post-quantum signature ML-DSA-65 and recorded in a verifiable audit trail. A blocked action leaves a forensic record of the attempt instead. Either way you keep provable evidence of what your agents tried to do, which makes compliance reviews and incident investigations far easier.
The plugin calls the Asqav cloud API directly from your Dify deployment. Only a minimal metadata bag is retained alongside a hash of the rest: action type, agent ID, session ID, model name, and tool name, in line with GDPR data minimization. If you prefer client-side hashing, you can run the Asqav Python SDK in hash-only mode alongside this plugin.
Setting up Asqav takes three short steps:
sk_.agent_.The Asqav plugin provides three actions for governing and proving agent activity.
Signs an agent action with ML-DSA-65. You provide an action type, such as read:data or tool:execute, plus optional context. You can also pass an optional action_ref to label the action so a pair of receipts can be linked later. The response includes an authorized flag so your workflow can branch on the decision instead of failing. An allowed action returns its signature, identifiers, timestamp, and a public verification URL. A blocked action returns a clear reason along with a signed denial receipt that is itself verifiable.
Verifies a signature by its ID. This is a public action that needs no authentication, so anyone can confirm that a given action was genuinely signed, who signed it, when, and with which algorithm.
Creates a multi-party signing session for high-risk actions. The action stays pending until enough approvers have signed off, which makes it a natural pre-execution gate for sensitive steps in a workflow.
Asqav fits into both Chatflow / Workflow apps and Agent apps:
authorized result so the workflow only proceeds when the action is permitted. To link a before-and-after pair of receipts for the same action, pass the same action_ref to both signs.Built by Asqav. Learn more at asqav.com.