AI LABS · AI ORCHESTRATOR

One control layer between the developer's coding tools and the models.

Developers keep VS Code with Copilot, Claude Code, Antigravity or Codex. The AI Orchestrator routes each task to a model, records usage, and applies policy, from one place. A platform that stays in place.

Every developer tool is wired to its own model, and the bill grows with adoption. Nobody can say which team, tool or task spends what, or which tools are unmanaged.

The AI Orchestrator gives the enterprise one control point: usage by harness, team and task, a routing policy per team, and governance evidence per run.

·THE CONTROL LAYER

Between the developer's coding tools and the models.

A plugin in each harness syncs the routing policy, routes each task and ships a run record to the governance console. Select a stage to see what it does.

Your developers' coding toolsAny harness · no tool changeVS Code + GitHub Copilot · Claude Code · Antigravity · Codex

AI Orchestrator

Route. Each task is routed through a connector to the model and provider the policy names. Planning, implementation, review, testing, documentation and debugging can each go to a different model.

ModelsAny model · any provider · any cloudGemini on Google Cloud · Anthropic · OpenAI · other clouds and provider endpoints

·WHAT IT RUNS

Four jobs, one layer.

The same four stages above, Connect, Route, Record and Govern, in depth. Model and infrastructure choices evolve without redesigning every harness integration, and usage is analysed by the unit that matters: developer activity and development task.

Connect

Harnesses

The plugin is installed in the developer's own harness: VS Code with GitHub Copilot, Claude Code, Antigravity or Codex. Developers keep their workflow. The plugin syncs the routing policy assigned to the developer, their team or the org.

Any harness, no tool changePolicy synced per developer, team or orgConnectors to the model and provider the policy namesOnboarding standardised without forcing one tool

Route

Task to model

Each task is routed to the model the policy names: planning, implementation, review, testing, documentation and debugging can each go to a different model and provider. Model and infrastructure choices change without redesigning the harness integration.

Task-to-model routing rulesApproved and blocked modelsFallback and exception pathsGemini as the default for the tasks it fits

Record

Usage

Every run is recorded with its harness, developer and team, task, model, provider, tokens and cost, and shipped to the governance console. Governed spend is shown beside unmanaged spend, and each cost states its basis: metered, estimated or unknown.

Runs by harness, team and developerCost by model, team and taskGoverned share of spendHarness native telemetry read beside the runs

Govern

Policy and evidence

The same layer is the policy boundary: approved model set per team, budgets and usage thresholds, workload restrictions, logging and exception handling. Policies are versioned and append-only; each version is the exact text the machines receive.

Versioned, append-only policiesBudgets and thresholds per teamPolicy decision captured per runException and override records

·THE ARCHITECTURE

How it plugs into your stack.

The harness starts the run. The orchestrator plugin resolves the execution path through a connector to the model the policy names. Telemetry records the run and ships it to the governance console.

Harnesses · plugin installed

VS Code + GitHub Copilottelemetry v2 run folder

Claude Codeplugin run folder · native telemetry

Antigravity · Codexrun folder · native telemetry

Tasks · routed by policy

plan · review · securityfrontier model

tests · debugGemini Flash

codegen · docsGemini Flash-Lite

AI Orchestrator plugin · on the developer's machinepolicy from the console · credentials stay local

Task with policy applied

Built-in connectorsprovider API · e.g. builtin-anthropic

MCP connectorsmodel dispatch · e.g. Gemini

Models

Gemini on Google Cloud

Anthropic

OpenAI

Other clouds and endpoints

Identity & secretsProvider credentials stay on the developer's machine. The console holds run metadata and policy text, and signs in with a provisioned email.

TelemetryRun records shipped to the console. Harness native telemetry read beside them, so unmanaged sessions are visible.

·THE ECONOMICS

We model the savings on real prices.

Routing turns model choice into a cost decision per task: the frontier model is reserved for judgment, and tests, code generation and documentation run on economical tiers. Two places let you model the economics on real prices and see the evidence.

AI Tokenomics ↗

Model it live.

The interactive policy router. Change the policy or the traffic mix and watch blended cost, escalation and quality recompute on real provider prices.

Open the router →Studies ↗

See the evidence.

The experiments behind the routing and tokenomics — what we tested across models and tiers, and what held up in production.

Read the Studies →

·HARD QUESTIONS

Common questions.

What buyers ask before they commit.

“Do our developers have to change tools?”

No. The plugin is installed in the harness they already use: VS Code with GitHub Copilot, Claude Code, Antigravity or Codex. The workflow stays the same. The policy decides which model each task goes to.

“Is this in the model-call path? What does it add?”

The plugin on the developer's machine resolves the model for each task and calls it through a connector. The governance console is not in the call path. It holds policies and run records, and a change reaches a machine only through the policy it syncs.

“How do you prove the savings?”

The console shows governed spend beside unmanaged spend, with each cost stating its basis. A routed policy is compared against the single-model baseline on the same work, and every run must pass build and tests before cost is compared. Published studies show 35% to 70% lower cost at quality parity.

Ask us a question→Engage via AI Orchestrator→