Interface: WebLlmConfig

Defined in: webllm.ts:14

Browser-only adapter backed by WebLLM (https://github.com/mlc-ai/web-llm). Models run on-device via WebGPU; no network for inference. The MLCEngine is loaded lazily on first stream so apps can ship the import without paying the wasm cost up front.

@mlc-ai/web-llm is an optional peer dependency — install it alongside this package when you opt into browser-only inference.

#Properties

#engine?

optional engine?: WebLlmEngineLike

Defined in: webllm.ts:21

Override the engine to inject a pre-loaded one (the MLCEngine spin-up is non-trivial — apps usually warm it once, not per turn).

---

#model

model: string

Defined in: webllm.ts:16

Model id from MLC's catalog, e.g. Llama-3.1-8B-Instruct-q4f16_1-MLC.

---

#onProgress?

optional onProgress?: (info) => void

Defined in: webllm.ts:23

Engine progress callback (model download / compile percent).

#Parameters

info

progress

number

text

string

#Returns

void

Explore nearby

• Peer
  AnthropicConfig

  Auto-generated API reference for AnthropicConfig.

• Peer
  AzureOpenAIConfig

  Auto-generated API reference for AzureOpenAIConfig.

• Peer
  BailConfig

  Auto-generated API reference for BailConfig.