Protocols
The open stack for AI‑native applications.
Seven protocols across three layers — rendering, context, and agents. Every spec is published openly, every implementation is conformance-tested, and every layer is proven in a product you can use today.
The stack
Three layers. One verifiable foundation.
Flagship spec
v2.0StableVirtual DOM Protocol
Structured UI representation that models understand natively. Generated UI becomes inspectable, diffable data — the foundation the rest of the rendering layer builds on.
- Validated component trees — every node is schema-checked before it renders
- Streaming layout patches — interfaces update incrementally mid-generation
- Stable node identity — documents stay diffable and cheap to compare
import { parseTree, applyPatch } from "@legionedge/vdom";
// Model output arrives as a validated VNode tree, not markup
const tree = parseTree(modelOutput);
for await (const patch of generation) {
applyPatch(tree, patch); // streaming layout, v2
}
export default tree.toReact();
Layer 01 · Rendering
What models draw.
Structured representations of interfaces and graphics that models read and write natively — and that machines can verify before a human sees them.
Proven in Nokuva- v2.0StableVirtual DOM ProtocolStructured UI representation that models understand natively.TypeScriptNode.jsRustSpec
- v1.4StableVirtual Browser ProtocolA simulated browser that gives models accurate rendering context.RustTypeScriptSpec
- v0.9BetaVirtual Canvas ProtocolA structured drawing surface for AI-generated graphics and layout.TypeScriptRustSpec
Layer 02 · Context
What models see.
How models see code, documents, and history: scoped, typed, and compressed so every token in the window is there for a reason.
Proven in Tavoc- v1.5StableModel Context ProtocolStructured context interfaces for optimal model comprehension.RustTypeScriptPythonSpec
- v0.9BetaContext CompressionReducing token usage while preserving model understanding.RustPythonSpec
- v1.2StableW0rkTree ProtocolEfficient source control and context management for agent-driven development.RustPythonSpec
Layer 03 · Agents
How models coordinate.
Hand-off, shared planning, and arbitration seams that stay reliable as the number of agents grows from one to a fleet.
Proven in Foltrac- v0.8BetaAgent ProtocolsCoordination and tool-use patterns for reliable multi-agent systems.PythonTypeScriptSpec
- Coordination benchmark suiteRuns against any conformant stack — shared-plan consistency, tool-use reliability under contention, and recovery when a collaborator fails mid-task. It exists because nothing off the shelf measured the thing that breaks.Benchmark report
Open by design
A protocol nobody can read is just an API.
Every spec ships with schemas, a conformance suite, and the research behind its design decisions. Independent implementations are the success condition, not a threat.