Skip to content
B BYOAg
Menu

Experimental patterns

BYOAg Patterns

BYOAg is being developed through small, composable interaction patterns rather than as a prematurely finalized universal protocol.

Each pattern describes a recurring way in which a user-owned agent can participate in an environment-defined experience.

These patterns are experimental. Their purpose is to make architectural questions explicit, support working prototypes, and establish a shared vocabulary for implementation and discussion.

Pattern 01

Agent Docking Bay

An application offers a user-friendly way to connect a personal agent without asking the user to understand APIs, URLs, or HTTP requests.

A Docking Bay can make the flow visible:

  • A ready-to-paste magic prompt starts the connection
  • The application shows when it is waiting for and has established a connection
  • A short human-entered pairing code confirms presence on both sides
  • The project declares a permission contract before remote capabilities are used

Conceptually, the flow is Project → Docking Bay → human pairing → permission contract → remote capabilities → operation, audit, and revocation.

Pattern 02

Prompt Package

An environment publishes a portable package containing the information an agent needs to participate in an experience.

A Prompt Package may include:

  • Context and objective
  • Available resources and constraints
  • Allowed actions and expected output
  • Evaluation criteria and interaction instructions

The package allows the environment to define the experience without requiring the user to abandon their preferred agent.

Pattern 03

Enriched Submission

An agent submits structured work to the environment. The response may include more than a success or failure status.

An enriched response may return:

  • Validation results and preliminary evaluation
  • Corrective feedback or missing information
  • Scoring and retry instructions
  • The next available action or a continuation of the experience

This turns BYOAg into an interaction model rather than merely a mechanism for distributing prompts.

Pattern 04

Continuation Chain

A submission may produce the next entry point in an experience, allowing the environment to reveal the next step only after evaluating the previous interaction.

Continuation Chains can support:

  • Guided workflows and educational exercises
  • Simulations and approval processes
  • Interactive stories and games
  • Multi-stage assessments

The environment maintains control of state and progression, while the user continues participating through their own agent.

Pattern 05

Agentic Contract and Evidence

An environment cannot assume that every agent will interpret or follow instructions in exactly the same way. Instead of relying on assumed compliance, the environment can declare:

  • Expectations, constraints, and acceptance criteria
  • Required evidence and permitted refusals
  • Partial-completion rules and observable outcomes

The agent may then report what it attempted, completed, refused, or could not verify.

Pattern 06

Asynchronous Continuation

Some experiences continue after the immediate interaction has ended. An environment may need to contact the user's agent when an external condition changes.

This may include:

  • An approval, cancellation, deadline, result, or status change
  • New availability or a request for intervention

This exploratory pattern considers authorized callbacks, subscriptions, event delivery, agent inboxes, resumable interaction state, stable agent identity, and safe context restoration. Its purpose is to allow event-driven interactions without requiring continuous polling.

Exploratory — not yet finalized

Status

Experimental and evolving

These patterns are experimental and are expected to evolve through prototypes, implementation experience, security analysis, and public discussion. They are not yet intended to represent a finalized standard.