Introducing AI Orchs & Orch Models

01What an AI Orch is

An AI Orch is the executable coordination object a system instantiates and operates. It binds the participants — models, agents, tools, APIs, workflows, memory and context, data, permissions, governance, runtime environments, compute, and execution receipts — together with their dependencies, state and execution logic, into one live operation. It exists at runtime: it can be created, modified, paused, resumed, forked, observed and completed. It is not a workflow with a new name; a workflow follows steps that were written in advance, and an AI Orch coordinates dynamically as conditions, outputs and policies change. Its receipts land on the orchestration graph and travel as A2F .orch documents.

Agents became a primitive. AI Orchs are the next one: not “we have an orchestration model,” but a new computational object.

02What an Orch Model is

An Orchestration Model is a model trained or designed to reason about coordination itself: which model should handle a task, which tool to invoke, what should occur in sequence and what in parallel, when to branch, when to retry, and how to manage state, permissions, cost, latency, dependencies and handoffs. Its output is an orchestration specification rather than prose about coordination: use model A for perception, call tool B, run agents C and D in parallel, require approval before action E, then reconcile the outputs. That specification, instantiated, is an AI Orch.

The relation is the one every runtime distinction in computing has needed. An LLM can generate the reasoning for a conversation, but the conversation is a separate runtime object with messages, state and participants. In exactly the same way: an Orch Model produces and governs; an AI Orch performs; and the AI Orch coordinates the agents, models and tools inside it.

Neither requires the other. A deterministic AI Orch — built from rules, dependency graphs and policy — runs today with no model controlling it. And an Orch Model need not execute anything: it emits the specification. The pairing is where the category gets interesting: an AI Orch becomes substantially more intelligent when an Orch Model is controlling and adapting it live.

03Where AI Orchs run today

None of this is promissory. LabsAI Studio runs on written AI Orchs now: the retrieval pipeline that covers a search with a spoken working-ack; the recovery ladder that reroutes a dying connection lane without ending the conversation; the shared-browser session with its own lifecycle and permissions; the media roster that queues, retires and hands off between three simultaneous streams. Each is a live coordination object with participants, dependencies, state, and pause/resume/observe semantics — deterministic, hand-written, and generating the trajectory ledger a learned orchestration layer trains against. The learning discipline for that step is the standing one: candidates are admitted only on held-out improvement through a gate, described in Larger Than Language §15.

04The taxonomy, stated once

Orchestration Intelligence is the domain — the intelligence required to organize intelligence. Orchestration Models (Orch Models) are the class that learns and reasons about orchestration; run from live speech, the class is Speech‑to‑Orchestration (STO). The Foundation Orchestration Model (FOM) is the general-purpose category. Octras — Orchestration-Centered Transducers for Routing and Agentic Systems — are the Labs model family implementing the class; in the terms that made the last decade legible, an Octra is a GOT, a Generative Orchestration Transducer. And AI Orchs are the executable orchestration structures the whole stack exists to produce, govern and improve. Beneath the runtime object sit the interfaces of the layer: the Orchestration Programming Interface (OPI) is how intelligence creates, governs and observes AI Orchs — its verbs are exactly this lifecycle — and the Parallel Agent Protocol (PAP) is how the coordinated work travels between participating intelligences. Spoken, the layer’s work arrives as Speech‑to‑Orchestration Voice Commands — the unit of spoken coordination an STO Model interprets — beside the Speech‑to‑Action Voice Commands the action side carries. One writing rule keeps the page honest at a glance: the runtime object is always “AI Orch,” and “Orch Models” always means the intelligence.