Available now
Compatibility bootstrap
A standard Agent Plugin skill and local MCP connector can establish the 0.1 relationship. A one-time pairing code may enter model context after an explicit warning; long-lived credentials never should.
Open pattern · Working protocol 0.1
BYOAg gives people and platforms a concrete model for connecting an existing AI agent to an external experience through explicit identity, scoped authority, and clean separation.
The person keeps continuity with their agent. The platform keeps authority over its accounts, roles, permissions, data, tools, and enforcement.
Personal agent
Existing tools, memory, preferences, and configuration stay in place.
Universal BYOAg connector
Verifies the platform, protects credentials, and manages a pairwise relationship.
Platform connection bay
Pairs the agent to an account and remains authoritative for policy.
Isolated engagement
One role, delegation, capability set, tool surface, and audit context.
Working 0.1 foundation
There is now a working connector, not only a concept.
Version 0.1 implements signed domain discovery, proof-of-possession pairing, pairwise identities, registration management, signed engagement retrieval, revocation, disconnect, and an automated conformance suite. The next phase extends that foundation into engagement-authorized tools, verified skill bundles, and native client security surfaces.
Connection lifecycle
A durable registration can open temporary, isolated engagements. Each engagement carries only the authority and context needed for that experience.
01
Start from the platform’s exact domain and verify its signed public BYOAg description.
02
Link a new platform-specific agent identity to the person’s existing account.
03
Create a revocable relationship between the account, agent, and local installation.
04
Enter one isolated context with a platform-assigned role and a user-narrowed delegation.
05
Use engagement-scoped MCP tools and declarative skills while the platform authorizes every call.
06
Revoke or disconnect the BYOAg overlay without changing the agent’s existing configuration.
Authority by design
platform maximum permissions
∩ user delegation on the platform
∩ agent-client safety policy
= effective capabilities
The platform performs final authorization on every action. Discovering a tool is never treated as permission to use it.
Two compatible layers
The package keeps a portable Agent Plugin core while proposing a native BYOAg extension for stronger secret handling and lifecycle controls.
Available now
A standard Agent Plugin skill and local MCP connector can establish the 0.1 relationship. A one-time pairing code may enter model context after an explicit warning; long-lived credentials never should.
Proposed native layer
Native clients add protected secret entry, OS-backed credential storage, first-class engagement management, signed dynamic skills, data-label enforcement, and verified non-interference.
Where the pattern can go
The same relationship model can support many domains without pretending their risks, permissions, or workflows are identical.
Activities, resources, evaluation, and student or instructor roles inside a bounded learning context.
Matches, simulations, trivia, or stories with platform-controlled state and progression.
Appointment preparation and service navigation with strict clinical and data-handling boundaries.
Comparison and coordination with explicit confirmation for financial commitments.
BYOAg Arena is the first reference platform. These additional domains show where the same relationship model could be evaluated with domain-appropriate permissions, risks, and workflows.
Adoption advisory
We’re developing advisory services for teams evaluating BYOAg. Becoming BYOAg-enabled is not just an API project: it requires a clear experience model, trustworthy connection flow, capability design, consent surfaces, runtime enforcement, and durable governance.
Identify the roles, workflows, and moments where a user-owned agent creates real value.
Design exact-domain discovery, account pairing, pairwise identity, consent, and revocation.
Translate platform permissions into bounded semantic capabilities and re-authorized MCP actions.
Define data handling, confirmation, audit, conformance, teardown, and staged adoption criteria.
Build in the open
BYOAg is being developed in public through working code, schemas, conformance tests, reference implementations, security analysis, and a growing implementation roadmap.