Permission-first planning

Know what your agent needs before it runs.

Choose its runtime, data, actions, credentials, retention, and approval policy. The builder calculates key and backend requirements locally.

Start building Browse templates

Local blueprint builder

Design the permissions before writing agent code.

Loaded from: Blank blueprint. Everything below stays in this browser tab unless you download or copy it.

Calculated operational riskLowThe current action set is read-only or produces drafts for review.

Generated configuration

My review agent

Low risk

AI API keyNot required

BackendNot required

External actionsNone

RetentionSession Only

Implementation notes

The current action set is read-only or produces drafts for review.
This design can remain device-local and does not require a Run Any AI backend.
A local model route can implement this design without an AI API key.
The selected approval policy keeps a person in the decision path.

{
  "schema_version": "runanyai.agent-blueprint.v1",
  "name": "My review agent",
  "objective": "Review user-provided material and produce a clear draft for approval.",
  "runtime": "local",
  "data": {
    "inputs": [
      "user_text"
    ],
    "retention": "session_only"
  },
  "actions": [
    "draft_response"
  ],
  "credentials": {
    "owner": "none",
    "secret_storage": "not_required"
  },
  "approvals": {
    "policy": "every_external_action",
    "external_actions_require_confirmation": true
  },
  "assessment": {
    "risk_level": "low",
    "backend_required": false,
    "api_key_required": false,
    "reasons": [
      "The current action set is read-only or produces drafts for review.",
      "This design can remain device-local and does not require a Run Any AI backend.",
      "A local model route can implement this design without an AI API key.",
      "The selected approval policy keeps a person in the decision path."
    ]
  }
}

This file is an architecture contract, not executable agent code. Review permissions, provider terms, costs, and security controls during implementation.

After export

A blueprint is the beginning of implementation.

Choose a model route. Local for small supported tasks; cloud for larger or shared workloads.

Implement least privilege. Request only the tools and data named in the contract.

Add approval gates. External and irreversible actions need clear confirmation and audit records.

Test failure modes. Validate with representative inputs before exposing the agent to real work.