DOCSAPI reference

INTERFACE

RLMAgentConfig

Defined in: packages/components/src/agents/rlm/RLMAgent.ts:19

RLMAgent configuration

Properties

Property Type Description Defined in
maxDepth? number Maximum nesting depth for RLM instances (default: 1) Remarks - depth=1: Root RLM → Base LLM (recommended for most tasks) - depth=2: Root RLM → Sub RLM → Base LLM (complex multi-layer analysis) - depth=3+: Multiple nested RLM layers (experimental) When maxDepth is greater than 1, nested RLM instances are created automatically. packages/components/src/agents/rlm/RLMAgent.ts:34
rootLLM LLMProvider LLM for generating orchestration code packages/components/src/agents/rlm/RLMAgent.ts:21
sharedContext? FileSystemContext Shared FileSystemContext for all recursion levels (optional) Remarks When provided, all nested RLM instances share the same FileSystemContext, avoiding redundant file reads through shared LRU cache. packages/components/src/agents/rlm/RLMAgent.ts:42
subLLM LLMProvider LLM for answering sub-queries packages/components/src/agents/rlm/RLMAgent.ts:23
systemPrompt? | string | SystemPromptBuilder Task-specific additional prompt (optional) Remarks This prompt is APPENDED to the default RLM system prompt. The default prompt describes REPL environment basics (tools, FINAL(), etc.). Your custom prompt should add task-specific strategies and examples. - String: Additional task-specific instructions - Function: Dynamic prompt generation based on context and query - Undefined: Uses only the default REPL environment prompt Example // Static task-specific prompt systemPrompt: "OOLONG-Pairs Strategy: Find all user pairs where both have entity/location..." // Dynamic prompt builder systemPrompt: (context, query) => Task-specific guidance for: ${query}`` packages/components/src/agents/rlm/RLMAgent.ts:64
verbose? boolean Mirror sandbox console output and errors to the real console. Default: false. packages/components/src/agents/rlm/RLMAgent.ts:68