Project status · September 2026
An open pattern with a working 0.1 foundation
BYOAg now has a working 0.1 protocol binding, a generic agent connector, a reference host, draft schemas, and tests. Important runtime, native-client, and production boundaries remain deliberately unfinished.
Current inventory
What exists and what comes next
Implemented at 0.1.0
Implemented- Signed, exact-domain discovery with same-origin JWKS validation
- Code-is-consent and platform-confirmation pairing modes
- Pairwise agent and installation identities with DPoP-bound credentials
- Registration listing and signed concurrent-engagement retrieval
- Registration revocation, disconnect, and local-state cleanup
- Six lifecycle MCP tools, a static bootstrap skill, and draft schemas
- An in-memory reference host and automated development conformance suite
Not implemented yet
Planned- Engagement-authorized remote MCP discovery, filtering, and invocation
- Signed declarative skill retrieval, verification, resolution, and activation
- OS-backed vault adapters and model-isolated native secret entry
- Data-label propagation and enforcement across tools and engagements
- JWKS rotation, DPoP nonce, DNS rebinding, and malicious-skill fixtures
- BYOAg Arena integration and independent security review
- Packaged plugin distribution and production operations guidance
Protocol and website
byoag
The public explanation, evolving information architecture, protocol documentation, examples, and production-static website.
Open the website repositoryReference implementation
byoag-agent-plugin
The universal bootstrap plugin, generic connector, reference host, draft schemas, protocol documents, and test suite.
Open the plugin repositoryRoadmap
From connection prototype to tested experience plane
- 01 · BUILT
Secure connection core
Discovery, pairing, identity, registration, engagements, and disconnect.
- 02 · NEXT
Experience tool plane
Capability resolution, tool filtering, invocation, confirmation, and audit.
- 03 · PLANNED
Verified skill lifecycle
Signed bundles, content digests, scoped activation, updates, and teardown.
- 04 · PLANNED
Native and production work
Secure client surfaces, Arena integration, hardening, and independent review.
Project principles
Constraints that should survive the implementation
User agency
The person keeps their agent and may narrow any proposed delegation.
Platform sovereignty
The environment owns its accounts, data, roles, tools, and final authorization.
Least privilege
Authority is explicit, contextual, revocable, and re-evaluated at runtime.
Pairwise identity
Cross-platform correlation is avoided by default rather than repaired later.
Evidence over assumptions
Behavior and conformance claims should be observable and testable.
Non-interference
Connecting and leaving must not disrupt the agent’s unrelated configuration.
Implementation-led learning
The draft evolves through prototypes, tests, and public discussion.
Honest status
Unimplemented and untested boundaries remain visible in the documentation.
Production readiness
A future claim that must be earned
A passing development suite is not enough. A production-readiness claim would require the missing native security surfaces, complete runtime enforcement, operational and abuse controls, broader adversarial tests, cross-implementation interoperability, deployment guidance, and independent security review.
Until then, BYOAg should be described as an experimental open pattern and implementation target—not a guaranteed interoperability layer.
Origin
An open exploration
BYOAg was conceived by Alejandro Gomez Flores as a way for people to bring their own AI agents into environment-defined experiences without surrendering the agent or the platform’s boundaries.
It is being developed openly through working prototypes, architectural exploration, implementation experiments, security analysis, and public discussion.