Back to Catalog

plugin

AI6 Agent Operating Layer

AI6 Agent Operating Layer is the controlled layer around agents that manages profiles, skills, scheduled work, hooks, sidecars, execution runs, and proof.

Host and data posture

Where this product is meant to live.

Internal AI6 execution rail; expose only customer-safe facades.

Commercial shape

Platform rail

Data posture

Internal only

Approval posture

Internal control plane

Best demo

StackFast UI

Best daily use

HermesStackFast UI

Who it is for

Enterprises building agent workflows that need approval, rollback, visibility, and proof.

Problem it solves

It prevents unmanaged agents, invisible scheduled work, and unsafe tool installation.

Activation model

Assisted setup with risk policy and approval gates.

Proof and source paths

  • Agent routes exist under /api/agent
  • Types and persistence exist
  • Enterprise/user guide docs exist

Repo evidence

  • app/api/agent
  • lib/agent-operating-layer
  • docs/AI6_AGENT_OPERATING_LAYER_ENTERPRISE_SPEC.md

Capabilities

skill lifecycle

plugin

Propose, approve, install, and rollback agent skills.

Agent use
Skill lifecycle management with approval gates.

Connector proof

Catalog is not the same as customer-ready.

This product only becomes market-ready after a real host mounts the entitled tools, a safe call succeeds, metering records the event, and a customer-safe receipt exists. Until then, the page shows the strongest proven status without rounding up.

Catalog

Runtime registry and product manifest are present.

Entitlement

Execution requires OAuth or storefront entitlement scope.

Callable

Callable tool proof is incomplete.

Metering

Metering waits on callable execution.

Customer validated

Pending fresh host-surface call receipt.

Runtime registry status

Primary registry

hybrid

Status

partial public mcp

Install gate

Assisted only until agent profile, skill lifecycle, cron, hook, rollback, and proof-ledger tools are exposed through one registry with smoke evidence.

Callable tools

  • stackfast.dispatch.create
  • stackfast.dispatch.claim
  • stackfast.dispatch.status
  • stackfast.dispatch.close
  • stackfast.project.news

Missing or gated tools

  • stackfast.agent.profile.resolve
  • stackfast.agent.skill.install
  • stackfast.agent.cron.schedule
  • stackfast.agent.hook.register
  • Storefront product is catalog-ready, not install-ready.
  • Current callable surface is the MCP communication/control-plane subset, not the full AI6 operating layer.

Install and activate

Connect this product through the StackFast MCP endpoint. Discovery metadata is public; execution uses OAuth and storefront entitlement checks.

MCP URL

https://stackfast.ai/api/mcp

Agent subscription

This product remains request-scope until its checkout, fulfillment, and public-name gates are complete.

Client snippets

ChatGPT

Add StackFast as a custom MCP connector, then approve OAuth when prompted.

Claude

Connect a remote MCP server using the StackFast MCP URL and a tenant-scoped token or OAuth flow.

Cursor

Register StackFast as a remote MCP server for project-scoped tool use.

Grok

Add StackFast as a custom MCP connector using the StackFast MCP URL and customer entitlement auth.

Product: AI6 Agent Operating Layer
Product API: https://stackfast.ai/api/store/products/agent-operating-layer
Pricing API: https://stackfast.ai/api/store/pricing/agent-operating-layer
Manifest API: https://stackfast.ai/api/store/manifests/agent-operating-layer
MCP URL: https://stackfast.ai/api/mcp
Auth: oauth_reader_or_storefront_entitlement
Checkout session API: https://stackfast.ai/api/store/checkout/session
Plans:
- Agent control scope: Request scope; checkout_ready=false; checkout_url=request_scope
Agent activation:
- Read this product manifest and pricing API.
- If a plan has checkout_ready=true, open its checkout_url or POST product_slug, plan_slug, and customer_email to the checkout session API.
- After Stripe payment, the signed webhook grants storefront_entitlements for the buyer email/account.
- Connect the LLM client to the MCP URL and approve OAuth so tools/list reflects the licensed reader surface.
- Call only tools exposed for that entitlement; unentitled execution must fail closed.

Agent-readable queries

enterprise agent operating layerAI agent skills hooks cron approval systemcontrolled AI agent workflows
What's on your mind?

Ask anything. Your thought enters the reasoning engine.

Enter to send · Shift+Enter for newline · Mic to speak