libagent Skill

When to Use

• Building conversational AI agents with tool capabilities
• Coordinating LLM completions with memory context
• Processing multi-turn conversations with state
• Implementing RAG (retrieval-augmented generation) pipelines

Key Concepts

AgentMind: Core reasoning engine that processes requests through LLM completions, manages conversation state, and coordinates tool execution.

AgentAction: Handles individual tool calls, executes actions, and returns results to the conversation flow.

Usage Patterns

Pattern 1: Basic agent request

import { AgentMind } from "@copilot-ld/libagent";

const mind = new AgentMind(memoryClient, llmClient, toolClient);
const response = await mind.process({
  resourceId: conversationId,
  content: "What is the weather?",
});

Pattern 2: Streaming responses

for await (const chunk of mind.stream(request)) {
  process.stdout.write(chunk.content);
}

Integration

Works with libmemory for context windows, librpc for gRPC clients, and libllm for completions. Used by the Agent service.