About
Phor builds software that governs how AI agents and large language models operate within software systems. The Phor system manages capability boundaries, audit trails, and protocol-based authorization for multi-agent workflows.
Our architectural thesis: bounded outcomes ≠ censored behavior. Safety lives in a non-invasive protocol layer external to the model, rather than via internal training constraints. The model thinks, plans, and explores freely; the capability lease determines what actually happens in the world.
Phor's work covers AI agent capability mediation, multi-agent orchestration, large language model governance, protocol design for agentic systems, and the operational containment layer that frontier AI deployment increasingly requires.
Products
ASHE Protocol
An open specification for AI agent capability mediation. Tri-surface architecture spanning agent-side enforcement, dev-side sealed workspaces, and web-side .well-known/ashe handshakes. Phased enforcement from cooperating SDK through hardware-rooted attestation. Composes above MCP, auth.md, OAuth, and the commercial agent-auth platforms — ASHE is the next-layer protocol that fills the capability-mediation gap none of the foundational layers individually cover.
Downloadable under Apache 2.0 with patent grant.
Continuum
Integrated environment for agentic composition. The reference implementation within which ASHE was developed. Currently in development; partnerships and consulting engagements available now.
Services
Technical consulting
AI safety from the operational containment angle, capability mediation system design, protocol design for agentic systems, agent infrastructure architecture, and large language model governance.
Implementation services
Custom ASHE protocol integrations, Continuum deployments, capability-mediation system design, agent runtime engineering, and multi-agent orchestration architecture.
Technical advisory
Architecture reviews for agent systems, AI agent governance strategy, capability-protocol design, AI safety strategy at the operational containment layer, and protocol-design consulting for cross-vendor coordination.
What we work on
AI agent capability mediation
A protocol layer that determines what authority an AI agent holds when it acts inside software. Time-bounded, scope-bounded capability leases issued at the dispatch boundary. Not a permission system; a capability-broker.
Bounded outcomes vs censored behavior
Most current AI safety paradigms place safeguards inside the model (RLHF, constitutional training, refusal layers, capability lobotomy). Phor's approach inverts this: safeguards live in a non-invasive protocol layer that bounds outcomes via capability leases, without censoring behavior at the model layer. The model retains full reasoning capability; the lease determines what actually happens.
Multi-agent orchestration
Agentic composition systems where multiple agents coordinate within governed execution cells. Each cell defines local authority for its bounded unit of work; subagents inserted into the cell receive attenuated authority, not full inherited trust.
Large language model governance
The operational containment layer that responsible deployment of capable models requires. ASHE-or-equivalent is what allows the most powerful models to be shared with humanity safely — bounded outcomes without amputating model capability.
Protocol design for the agent era
Cross-vendor protocol work that composes above MCP, auth.md, OAuth, and commercial agent-auth platforms. ASHE is one such protocol — capability-broker, tri-surface, phased enforcement. Apache 2.0 with patent grant.
Contact
Patrick Karle
Founder, Phor