Treat it as a signal, not something you can install. Sierra says a v0.1 specification is due later in October, so for now it shows where brands and personal agents are heading. Businesses running a Sierra or Meta agent should follow it, and agent builders should wait for the spec.
What changed
- · Release Sierra and Meta announce Personal Agent Protocol source
Personal Agent Protocol is an open standard announced on 6 October 2026 by Sierra and Meta, with Genesys, Instinct, Rocket, Shopify, Stripe and Walmart named as partners. It defines how a consumer's personal AI agent starts a session with a business, signs in over OAuth, and acts through the company's website, APIs or its own agent.
Maker: Sierra and Meta · Protocol: CUSTOM · Auth: oauth
Compatible agents: Personal AI agents acting for consumers (no specific agent confirmed as supporting it)
Required runtime: None published: Sierra plans a v0.1 specification and a reference implementation later in October 2026
About Personal Agent Protocol
Personal Agent Protocol (PAP) is an open standard that Sierra and Meta announced on 6 October 2026. It describes how a consumer's personal AI agent starts a session with a business, shows who it acts for, and finishes a job through the channel the company picks. It is a skill in HokAI's sense, a capability agents plug into, rather than a product a person opens. Bret Taylor and Clay Bavor of Sierra wrote the launch post, and Meta is co-developing the standard. The post names six partners: Genesys, Instinct, Rocket, Shopify, Stripe and Walmart.
As Sierra describes it, a session starts on the company's website, where the agent finds out what the business offers and how to reach it. The agent can begin as a guest, which may be enough to check stock or ask about a returns policy. When a task needs account access, the customer signs in on the company's page or uses credentials already set up with their agent, and decides whether the agent gets read-only or write access. The session is built on OAuth and carries across channels, so a question asked before sign-in and an order change made afterward belong to the same visit.
The company chooses the route the agent takes: its regular web pages, its APIs built on standards such as MCP and OpenAPI, or its own agent for jobs that need conversation, such as a warranty claim. The business sets what each route allows and gets visibility into what personal agents do. The launch post does not name any personal agent that already supports the standard. Agents of the kind it targets include Muse, OpenAI dots, Gemini Spark, Grok Bot, Cue, ChatGPT Work, OpenClaw, Hermes Agent and Manus, but HokAI found no statement that any of them implements it.
The post calls the standard open for anyone to implement and lists no price, fee or licence terms. Sierra says it plans to publish a v0.1 specification later in October 2026, host design workshops and release a reference implementation. Payments, finer-grained permissions and push notifications are named as possible later extensions, not part of the first release. No country availability is stated.
Treat PAP as an early signal. Until a specification and reference code are public, what it will do is vendor intent attributed to Sierra and its partners, and the details can change before v0.1 lands.
Key Features
- Guest-first sessions: An agent can start without signing in, enough for stock checks or a returns question, and sign in only when a task needs the customer's account.
- Customer-chosen access: Once signed in, the customer sets the agent to read-only or lets it make changes.
- OAuth session across channels: OAuth underpins the session, and it stays one visit across channels, from the first guest question to a later order edit.
- Three routes for the company: A business can expose its website, its APIs built on MCP or OpenAPI, or its own agent.
- Company-set rules and visibility: The business decides what personal agents may do on each route and sees when one is acting for a customer.
- Planned extensions: Sierra names finer permissions, push notifications such as flight delays, and payments as later additions.
Use Cases
- Check stock or a returns policy as a guest: A personal agent opens a session on a retailer's site without signing in and reads availability or return rules, which the launch post says may be enough for simple questions.
- Change an order after signing in: The customer signs in and grants write access, and the agent edits the order inside the same session that began with the earlier guest question.
- File a warranty claim through the company's own agent: For a job that needs conversation, the personal agent hands the task to the business's agent while the customer's access settings still apply.
Requirements
- Nothing to install yet: the v0.1 specification and a reference implementation were announced as coming later in October 2026
- Businesses need a website or APIs where OAuth sign-in and customer-chosen read or write access can be offered
How to Invoke
No callable interface is published yet. As described, an agent starts a guest session on the company's site, then continues over OAuth and uses the company's website, APIs (MCP or OpenAPI) or its own agent.
Pricing
No price, fee or licence terms are listed. Sierra's launch post (read 25 October 2026) calls the standard open for anyone to implement. Costs fall on whoever builds support for it.
Strengths
- Six named partners at launch (Genesys, Instinct, Rocket, Shopify, Stripe, Walmart) give it more backing than a single-vendor draft.
- Reuses OAuth, MCP and OpenAPI, so a company with existing APIs does not start from zero.
- Customers pick read-only or write access, which keeps the person in control of what the agent can change.
- Open for anyone to implement, so a small agent builder is not locked out by a fee.
Weaknesses
- The v0.1 specification was still described as upcoming on the pages HokAI read, so there is nothing to build against yet.
- No personal agent is confirmed as supporting it, so a business gains no traffic from implementing it today.
- Payments are listed as a possible future extension, so purchases are outside the first release.
- Licence terms and governance of the standard are not stated, so openness is a claim to verify once the spec ships.
Frequently Asked Questions
How much does Personal Agent Protocol cost in 2026?
Sierra's launch post calls it open for anyone to implement and lists no price, fee or licence terms. The cost therefore sits with each company or agent builder: engineering time to support OAuth sessions and whichever routes they expose. The pages HokAI read on 25 October 2026 did not include a published specification, so licence terms are unconfirmed.
Is Personal Agent Protocol free to use?
The launch post mentions no charge for implementing it. It is a specification rather than a hosted service, so there is nothing to subscribe to. What a business pays for is the products built on it, and those prices are set by their makers.
Which tools compete with Personal Agent Protocol in 2026?
It has no installable rival today. The closest overlap is agentic commerce work such as the Trusted Agent Protocol tied to Visa, which The Next Web reports Stripe and Shopify had already joined. MCP and OpenAPI are not competitors: Sierra names them as ways a company can expose its APIs underneath PAP.
What separates Personal Agent Protocol from MCP?
MCP connects an agent to tools and data. PAP covers the step before that, as Sierra describes it: how a consumer's agent identifies itself to a business, starts as a guest, and gets read-only or write access chosen by the customer. The two stack, since a company's APIs under PAP can be built on MCP. MCP already has public specification documents, while PAP's v0.1 was still upcoming in the launch post.
How do you get started with Personal Agent Protocol?
There is nothing to install yet. Sierra says it plans a v0.1 specification, design workshops and a reference implementation, so the first step is to read the spec when it lands and join a workshop. A business can prepare by deciding which routes to expose (website, APIs on MCP or OpenAPI, or a company agent) and what read-only and write access should mean for its customers.