ModelRunnerBackendConfig

⚠️ SECURITY WARNING: Enabling browser mode may expose API keys in client-side code. This option should ONLY be used for development and testing. Production applications should always use proxy servers to protect API keys.

Each provider implements browser compatibility differently:

Anthropic: Adds anthropic-dangerous-direct-browser-access: true header
Gemini: Already browser-compatible (API key in URL), this flag has no effect
OpenAI: Already browser-compatible, this flag has no effect
Other providers: May have provider-specific implementations

Default

false

Example

// Development only - DO NOT use in production!
const backend = new AnthropicBackendAdapter({
  apiKey: process.env.ANTHROPIC_API_KEY,
  browserMode: true  // ⚠️ Exposes API key in browser
});

Inherited from

BackendAdapterConfig.browserMode

cacheModels?

readonly optional cacheModels: boolean

Defined in: packages/ai.matey.types/src/adapters.ts:223

Enable model list caching.

Default

true

Inherited from

BackendAdapterConfig.cacheModels

communication

communication: CommunicationConfig

Defined in: packages/ai.matey.types/src/model-runner.ts:271

Communication configuration.

custom?

readonly optional custom: Record<string, unknown>

Defined in: packages/ai.matey.types/src/adapters.ts:164

Provider-specific configuration options.

Inherited from

BackendAdapterConfig.custom

debug?

readonly optional debug: boolean

Defined in: packages/ai.matey.types/src/adapters.ts:154

Enable debug logging.

Default

false

Inherited from

BackendAdapterConfig.debug

defaultModel?

readonly optional defaultModel: string

Defined in: packages/ai.matey.types/src/adapters.ts:199

Default model to use when no model is specified in the request. This provides a fallback model for requests that don’t specify one.

Example

'gpt-4o' for OpenAI, 'claude-3-5-sonnet-20241022' for Anthropic

Inherited from

BackendAdapterConfig.defaultModel

headers?

readonly optional headers: Record<string, string>

Defined in: packages/ai.matey.types/src/adapters.ts:159

Custom HTTP headers to include in requests.

Inherited from

BackendAdapterConfig.headers

lifecycle?

optional lifecycle: LifecycleConfig

Defined in: packages/ai.matey.types/src/model-runner.ts:281

Lifecycle management configuration.

maxRetries?

readonly optional maxRetries: number

Defined in: packages/ai.matey.types/src/adapters.ts:148

Maximum number of retries for transient failures.

Default

Inherited from

BackendAdapterConfig.maxRetries

model

model: string | ModelReference

Defined in: packages/ai.matey.types/src/model-runner.ts:261

Model configuration.

models?

readonly optional models: readonly (string | AIModel)[]

Defined in: packages/ai.matey.types/src/adapters.ts:211

Static model list (used when provider doesn’t have listing endpoint or to override remote list).

Can be either:

Array of model IDs (strings) - will be normalized to AIModel objects
Array of full AIModel objects with capabilities

Inherited from

BackendAdapterConfig.models

modelsCacheScope?

readonly optional modelsCacheScope: "global" | "instance"

Defined in: packages/ai.matey.types/src/adapters.ts:237

Cache scope strategy.

‘global’: Share cache across all adapter instances (default)
‘instance’: Each adapter instance has its own cache

Default

'global'

Inherited from

BackendAdapterConfig.modelsCacheScope

modelsCacheTTL?

readonly optional modelsCacheTTL: number

Defined in: packages/ai.matey.types/src/adapters.ts:229

Cache TTL in milliseconds.

Default

3600000 (1 hour)

Inherited from

BackendAdapterConfig.modelsCacheTTL

modelsEndpoint?

readonly optional modelsEndpoint: string

Defined in: packages/ai.matey.types/src/adapters.ts:217

URL endpoint for fetching models (overrides default). Used for custom model endpoints or proxies.

Inherited from

BackendAdapterConfig.modelsEndpoint

port?

optional port: number

Defined in: packages/ai.matey.types/src/model-runner.ts:287

Port number for HTTP communication. If not specified, an available port will be auto-discovered.

process

process: ProcessConfig

Defined in: packages/ai.matey.types/src/model-runner.ts:266

Process execution configuration.

promptTemplate?

optional promptTemplate: string | PromptTemplate

Defined in: packages/ai.matey.types/src/model-runner.ts:293

Prompt template name or custom template.

Example

'llama2', 'chatml', 'alpaca'

runtime?

optional runtime: RuntimeConfig

Defined in: packages/ai.matey.types/src/model-runner.ts:276

Runtime configuration for model execution.

streaming?

readonly optional streaming: StreamingConfig

Defined in: packages/ai.matey.types/src/adapters.ts:251

Streaming configuration for this backend.

Controls how streaming responses are delivered:

mode: ‘delta’ (incremental only) or ‘accumulated’ (full text each chunk)
includeBoth: Whether to provide both delta and accumulated in chunks
bufferStrategy: How to buffer for accumulated mode

Default

{ mode: 'delta', includeBoth: false, bufferStrategy: 'memory' }

Inherited from

BackendAdapterConfig.streaming

timeout?

readonly optional timeout: number

Defined in: packages/ai.matey.types/src/adapters.ts:142

Request timeout in milliseconds.

Default

Inherited from

BackendAdapterConfig.timeout