November 17, 2026 – Master of Code Global has made their proprietary LOFT, an LLM-Orchestrator Framework for TypeScript, public. The framework runs the LLM side of a chat backend: sessions, chat history, prompts, middleware, tool calls, and rate limits. It is free to use, change, and ship under the MIT license, and the code, tests, and security policy are open to read in the LOFT repository on GitHub.
What Is LOFT?
LOFT runs inside the backend a team already has, such as Express, NestJS, or AWS Lambda, so sessions and history stay in the client’s own database and cloud. A route calls one method, and LOFT runs the rest of the turn. As of November 2026, the repository ships 13 packages, published on npm under @moc-global.
The framework covers the parts that turn a chat prototype into a product:
- Model providers. OpenAI, Azure OpenAI, Anthropic Claude, and Google Gemini share one message format, and a conversation can switch models between turns.
- Storage. Sessions and chat history go to PostgreSQL or a custom store. System prompts and model settings live as JSON manifests on local disk, AWS S3, Google Cloud Storage, or Azure Blob Storage, and reload without a redeploy.
- Rate control. Each model gets its own queue, token-bucket rate limiter, and retries. A default cap of 10 model calls per request stops a looping agent.
- Agents and tools. LOFT connects to any Model Context Protocol (MCP) server. The loft-agents package adds agent graphs, parallel fan-out, and teams.
- Tracing. Every middleware step, model call, and tool call can be sent to Langfuse or a custom tracer.
LOFT focuses on conversational products and small- to mid-size agentic workflows.
Why Is Master of Code Global Publishing LOFT?
Master of Code Global has built conversational AI since 2016, and LOFT came out of that delivery work. It is the framework the company uses for most of its Conversational AI and AI agent development projects. According to figures from projects delivered on LOFT, it cut initial setup effort by 43%, saved up to 20% of the budget when scaling before MVP, and made project support 3x faster.
We built LOFT because chat features keep asking for the same parts: memory, a second model, prompt edits, rate limits, and tools. Building them once and testing them means a client’s team starts from working components instead of an empty repository. For clients, that means a faster start, a switch between AI models without rewriting code, prompt changes that don’t wait for a release, and a system that runs on their own cloud and their own AI provider contracts.
With the repository public, clients and their engineers can read how the framework works before any contract is signed.
The MIT license also means no lock-in. LOFT runs on the client’s cloud and AI provider contracts, with no Master of Code Global server in the middle, and the team owns the code it runs.
How Can Businesses Check LOFT for Themselves?
Businesses can review LOFT from their own side before any conversation with us. Their engineers can read the code, tests, and security policy on GitHub and run the quickstart in about 15 minutes. It needs Node.js 22 or later, a PostgreSQL database, and a JSON manifest.
Read the code on GitHub, or talk to our LOFT engineers about your project.
About Master of Code Global
Master of Code Global is an AI consulting and engineering company that helps organizations move from AI strategy to implementation. Its services cover readiness assessment, architecture, development, integration, deployment, training, and optimization.

