SUVENRAAI Works Within Your Rules
HelpDevelopers

Local-first AI authorization

Let AI operate within clear permissions

SUVENRA is an AI task authorization, handoff, and review system.

Create AI taskConnected Agent
  • One-time authorization works without an account
  • Signing keys are decrypted only on your device; signed-in users can sync end-to-end encrypted records
  • Human approval remains the final authority

Workflow

Choose how you want to start

One-time task · Complete example

Compare Notion, Obsidian, and Google Docs as a knowledge base for a 10-person team, then recommend one within 24 hours

Connected Agent

Pair once, then deliver future authorizations directly to this Agent.

  1. 1
    Task and evidence

    Gather sources and return a cited summary for approval.

  2. 2
    Copy AI handoff

    Work only within this signed authorization. Return the result and supporting evidence for human review; the authorization is not final approval.

  3. 3
    Review returned work

    Compare the returned work with the signed task and required evidence, then record your decision.

When to use SUVENRA

See what SUVENRA adds before an Agent acts

The task may be ordinary. SUVENRA adds independent, signed limits, evidence requirements, expiry, and human review.

When to use SUVENRA

Complete example · Research agent

Compare Notion, Obsidian, and Google Docs as a knowledge base for a 10-person team, then recommend one within 24 hours
SUVENRA · Set clear limits for one AI task

Define the task, allowed and blocked actions, required evidence, and expiry. Sign it, then share the least-privileged AI link.

Agent identity
Research agent
Expires
24 hours
Task
Compare Notion, Obsidian, and Google Docs as a knowledge base for a 10-person team, then recommend one within 24 hours
Expected evidence
Use at least 3 official sources. Include URLs, access times, a feature comparison, and clearly separate facts from recommendations
AI can
Read and researchPrepare draftsReturn evidence
Guardrails
No spendingNo publishingHuman approval required
Result summary
Return a concise comparison table, a recommendation with reasons, and a source list. Do not publish or spend money
  1. 1
    Create AI task

    Compare Notion, Obsidian, and Google Docs as a knowledge base for a 10-person team, then recommend one within 24 hours

  2. 2
    Sign authorization → Copy AI link

    Work only within this signed authorization. Return the result and supporting evidence for human review; the authorization is not final approval.

  3. 3
    Agent view: Accept this task → Submit work for review

    Return a concise comparison table, a recommendation with reasons, and a source list. Do not publish or spend money

  4. 4
    Review returned work

    Compare the returned work with the signed task and required evidence, then record your decision.

SUVENRA Agent Protocol 1.4

One integration, reusable authorization

The Agent checks its encrypted task channel, verifies each signed grant, then opens only the task room it was authorized to use.
1Sign authorizationSign authorizationAn explicit allowlist for the Agent. Any action that is not selected remains outside the authorization.Read full guidance
2Agent task channelAgent task channelAn encrypted pull endpoint for signed tasks. The Agent must install its credentials and check this endpoint; creating a SUVENRA connection alone does not make the Agent receive anything.Read full guidance
3Verify authorizationVerify authority and evidence before trusting the resultChecks the signature, controller, service namespace, scope, expiry, required fields, and revocation status. It does not prove that an Agent's factual claims are true.Read full guidance
4Review returned workReview returned workA person accepts or rejects the returned work. SUVENRA binds that decision, the original authorization, and the evidence into a signed outcome receipt.Read full guidance
Tasks, room capabilities, and delivery keys remain encrypted
Create a connection

What SUVENRA proves

Signing keys are decrypted only on your device; signed-in users can sync end-to-end encrypted records

Human approval remains the final authority

Help centerRevoke authorization