Agent Tool Dispatch

import requests def dispatch_search(intent: str, query: str) -> dict: platform_map = {"product": "amazon", "video": "youtube", "discussion": "reddit", "general": "google"} platform = platform_map.get(intent, "google") res = requests.post( "https://api.scavio.dev/api/v1/search", headers={"x-api-key": "YOUR_KEY"}, json={"query": query, "platform": platform}, ) return res.json()

Agent Tool Dispatch is relevant to Google, Amazon, YouTube, TikTok, Reddit. Scavio provides a unified API to access data from all of these platforms.

Agent tool dispatch is the core runtime decision loop in any tool-using AI agent. When a user asks 'what are the top results for X,' the agent must: (1) recognize this requires a search tool, (2) select which search tool (Google, Amazon, Reddit, etc.), (3) construct the correct API call with parameters, (4) execute the call and handle errors, (5) parse the response into conversational context. Dispatch patterns in 2026: Function calling (OpenAI/Anthropic native) -- the LLM outputs a structured tool call that the runtime executes. MCP (Model Context Protocol) -- tools are discovered dynamically and called through a standardized protocol. ReAct loop -- the agent reasons about which tool to call, acts, observes the result, and repeats. Plain code -- the agent writes and executes Python/JS code that calls APIs directly. Cost-aware dispatch: at $0.005/query (Scavio) or $0.008/credit (Tavily), tool calls have real cost. Smart dispatch avoids redundant calls: cache recent results, batch related queries, and skip searches when the answer is already in context. A poorly designed dispatch loop can make 5-10 search calls per user question. A well-designed one averages 1-2. Latency budget: each tool call adds 200-2000ms of latency. Dispatch design must balance thoroughness (more calls = better answers) against responsiveness (users expect sub-3-second responses). Parallel dispatch (calling multiple tools simultaneously) reduces wall-clock time at the cost of higher API spend.