LLM Providers

Loopstack supports multiple LLM providers through a runtime registry. Provider modules self-register at startup. Workflows and tools resolve providers by name — swap or use multiple providers in parallel without changing workflow code.

Quick Start

Import a provider module. That’s it — the adapter tools are available globally.

import { ClaudeModule } from '@loopstack/claude-module';
 
@Module({
  imports: [LoopCoreModule, ClaudeModule],
})
export class AppModule {}

Tool Defaults with @InjectTool

Configure provider and model at the injection site. Call sites stay clean — only pass prompt, tools, and messages.

@Workspace({ ... })
export class MyWorkspace implements WorkspaceInterface {
  @InjectTool({ provider: 'claude', model: 'claude-opus-4-7', tools: ['read', 'glob'] })
  llmGenerateText: LlmGenerateTextTool;
 
  @InjectTool({ provider: 'claude' })
  llmDelegateToolCalls: LlmDelegateToolCallsTool;
}

// Workflow — only call-specific args
const result = await this.llmGenerateText.call({
  system: 'You are a helpful assistant.',
  messagesSearchTag: 'message',
  // provider, model, and tools come from @InjectTool defaults
});

Defaults are deep-merged with call-site args. Explicit call-site values always win. undefined does not override a default.

@InjectWorkflow Defaults

Same mechanism for sub-workflows. Defaults merge into every run() call before BullMQ serialization:

@InjectWorkflow({ provider: 'claude', model: 'claude-opus-4-7' })
chatAgent: ChatAgentWorkflow;
 
// Defaults are included automatically
await this.chatAgent.run({ system, tools, userMessage });

Using Multiple Providers

Import both modules and configure each tool or workflow injection with its provider:

@Module({
  imports: [LoopCoreModule, ClaudeModule, OpenAiModule],
})
export class AppModule {}
 
@Workspace({ ... })
export class MyWorkspace implements WorkspaceInterface {
  @InjectWorkflow({ provider: 'claude', model: 'claude-opus-4-7' })
  smartAgent: ChatAgentWorkflow;
 
  @InjectWorkflow({ provider: 'openai', model: 'gpt-4o-mini' })
  fastAgent: ChatAgentWorkflow;
}

Both agents use the same ChatAgentWorkflow class. The provider is selected by the provider arg.

Adapter Tools

All LLM interactions go through adapter tools from @loopstack/llm-provider-module. This ensures validation, interceptors, and logging apply to every LLM call.

Message Documents

All providers share a single LlmMessageDocument with normalized content. Native API responses are stored in entity.meta.response for provider-specific round-trips.

Environment Variables

Prefer @InjectTool defaults over env vars in production.

Available Providers