Lloyal

Product Introduction

1. Definition: Lloyal is a TypeScript-based development platform and runtime for building and deploying vertically integrated AI applications. It is a Vertical Inference platform that combines the application logic, multi-agent orchestration, and the live language model inference state into a single, programmable system.
2. Core Value Proposition: It exists to enable developers to ship AI applications that think and act as standalone, offline-capable binaries or services, eliminating the complexity of managing separate inference servers, API keys, vector databases, and orchestration frameworks like LangGraph.

Main Features

1. Harness Development Kit (HDK): A TypeScript SDK that scaffolds a complete, working AI application with one command (npx lloyal-ai new). It includes a built-in model runtime, agent system, and retrieval capabilities. The harness acts as the application's controller, written in standard TypeScript with Effection for concurrency, governing how agents collaborate and act.
2. Abilities Ecosystem: A secure channel for 1st and 3rd party extensions that go beyond standard tool calling. Abilities are inference-native, meaning they execute within the model's own runtime context. This allows them to access the calling agent's live state, knowledge, and lineage, and even fork execution into new agents or require specific models to be loaded.
3. Vertical Inference Architecture: This is the core architectural paradigm. It unifies the application (Views), the control logic (Controller/Harness), and the resident LLM (Model) into one cohesive stack. The live model's context and multiple agent branches are managed as a shared resource, allowing features like single-model dispatch for multiple concurrent agents, reducing compute overhead.

Problems Solved

1. Pain Point: The excessive fragmentation and operational complexity in building production AI apps. Developers typically need to wire together separate services for inference (OpenAI API, Ollama), orchestration (LangChain, LangGraph), vector databases, and backend logic, leading to latency, high costs, and deployment headaches.
2. Target Audience: TypeScript/JavaScript developers building commercial or enterprise AI applications; product teams needing to ship embedded, offline AI features; startups focusing on vertical AI solutions who want to own their full stack without vendor lock-in.
3. Use Cases: Shipping a diagnostic assistant as a desktop app for field technicians without internet; deploying a multi-agent research tool as a private appliance on-premise; creating a cross-platform (CLI, Web, Desktop) AI co-pilot where the core intelligence logic is identical across all surfaces.

Unique Advantages

1. Differentiation: Unlike meta-frameworks that compose external services, Lloyal provides an integrated, batteries-included runtime. It contrasts with using OpenAI's API + LangGraph by offering offline execution, no per-call costs, and deeper integration. It differs from local servers like Ollama by baking orchestration and application logic directly into the runtime.
2. Key Innovation: The Abilities system represents a significant leap from Level 0 (tool calling) and Level 1 (MCP) protocols. By granting extensions access to the live inference state and the authority to spawn agents, it enables more powerful, context-aware, and composable agent behaviors natively. The concept of a shared KV context across multiple agents for efficient compute utilization is also a key technical innovation.

Frequently Asked Questions (FAQ)

1. What is Vertical Inference? Vertical Inference is an application architecture where the AI model's runtime and live reasoning state are directly integrated into the application's own process and control flow, managed by standard application code (like TypeScript) rather than being a separate, remote service.
2. How does Lloyal handle different AI models? The Lloyal harness is model-agnostic. The platform allows you to define which model your application uses, from small on-device models (e.g., Qwen 2.5 4B) to large frontier models (e.g., GLM-5.2). The same application code and Abilities can work across different model placements.
3. Can I use Lloyal to build a web application? Yes. Lloyal applications can be shipped to multiple fronts: as a Terminal CLI, a Desktop app (via Electron/Tauri), a Web application, and soon Mobile. The core harness (business logic and AI orchestration) remains the same across all these surfaces.
4. Is an internet connection or API key required? No. A primary feature of Lloyal is enabling fully offline AI applications. Once the model is bundled with your application binary, it runs independently without needing external API keys or cloud inference services.
5. What is the difference between an Agent and an Ability in Lloyal? An Agent is an instance of a reasoning process spawned by the harness. An Ability is a capability (tool, data source, skill, or even a model) that an agent can use. Abilities are installed into the harness and become available for all agents to utilize within their execution context.