AiMatch

Lloyal

Lloyal gives TypeScript developers a scaffold and runtime for building AI applications with programmable agents and live model inference. Teams can deploy the same application offline on-device, on a private appliance, or on frontier compute.

One of 576 tools in AI Agents

Visit Lloyal Compare alternatives Own this tool? Claim it
Lloyal screenshot

Who it's for

  • AI application developers building a TypeScript harness to control model and agent behavior.
  • Engineering teams that need to deploy an AI application on-device and offline or as a service on a private appliance.
  • Teams building multi-agent applications that need extensions to access live inference state and fork it into new agents.

Not the right fit if…

  • Teams planning to ship an iOS or Android app immediately: the page lists mobile support as “soon.”

How it fits your workflow

Run the CLI scaffold command, then choose the application’s front ends, model, and template. The generated application includes the model, agents, retrieval, and runtime, and can be changed in TypeScript.

Write the harness to govern agent tasks and application behavior. The page’s incident example runs research tasks in parallel, reconciles their evidence, and can pause for an operator decision.

Deploy the application as an offline on-device binary, a multi-user service on a private appliance, or online on frontier compute. The page says no OpenAI or Anthropic key, Docker, Ollama, LangGraph, or vector database is required for the scaffolded setup.

Pricing

The vendor doesn't publish prices. Check with Lloyal directly.

Prices checked on Oct 2, 2026 from the vendor's site. They can change; confirm before you buy.

Key features

TypeScript application scaffolding
The `npx lloyal-ai new` command creates a working application. Developers choose its front ends, model, and starting template, then modify the generated code.
Flexible deployment
The page describes shipping an application as a single on-device binary that runs offline, as a multi-user service on a private appliance, or online on frontier compute.
Inference-native Abilities
Abilities run inside model execution rather than behind a network boundary. They can access information about the calling agent and live inference state, and can fork state into new agents.
Signed Abilities channel
Lloyal hosts a signed channel of first- and third-party Abilities. Installing an Ability brings its tools and schemas, a live source, situation-aware skills, and composed models into an application.
Programmable agent harness
Ordinary TypeScript code governs how an application’s intelligence retrieves, reasons, acts, recovers, and continues. The page shows an incident harness that runs multiple agent tasks and reconciles their evidence.
Local and frontier model examples
The page demonstrates an on-device Qwen 3.5 4B setup and a hosted GLM-5.2 setup across two B200 GPUs, using the same application program.

Limitations to know

  • Mobile support for iOS and Android is marked as “soon” on the page; availability is not described as released.

Getting started

npx lloyal-ai new

Build the harness in TypeScript. The page says no OpenAI or Anthropic key, Docker, Ollama, LangGraph, or vector database is required.

  1. Run `npx lloyal-ai new` to create a working application scaffold.
  2. Choose the front ends, model, and template.
  3. Change the generated application and harness to implement the desired behavior.