--- title: "Tool Calling" description: "4 posts about Tool Calling, summarised, each linking to the original." --- # Tool Calling > 4 posts about Tool Calling, summarised, each linking to the original. ## Articles ### [How to Build a Voice AI Agent with OpenAI's GPT-Live, Twilio Agent Connect, and TypeScript](https://yomu.fyi/post/how-to-build-a-voice-ai-agent-with-openai-s-gpt-live-twilio-agent-conn-typescript.md) - Company: [Twilio](https://yomu.fyi/company/twilio.md) - Author: Wen Zhu, Xinghao Huang - Published: Oct 15, 2026 - Type: Tutorial The post addresses the common problem of voice AI demos failing in production by introducing a solution using OpenAI’s GPT-Live and Twilio Agent Connect. GPT-Live is a full-duplex speech-to-speech model that separates conversation from background work, and Twilio Agent Connect provides a GPTLiveProvider to bridge Twilio Media Streams to GPT-Live without custom WebSocket plumbing. The tutorial walks through building a TypeScript application that connects inbound callers to a GPT-Live-powered voice agent, including setting up a Fastify server, configuring the session with a welcome instruction, and wiring tool calls via Responses delegation using a get\_weather example. It also covers placing outbound calls using the same provider and demonstrates features like per-call customization and transcript logging. The result is a production-ready voice agent that can greet callers, answer questions, and execute backend functions, with the same pattern applicable to scenarios like appointment booking or order status lookups. ### [Plan or Retrieve First? A Scale Study of Tool Discovery Architectures](https://yomu.fyi/post/plan-or-retrieve-first-a-scale-study-of-tool-discovery-architectures.md) - Company: [Apple](https://yomu.fyi/company/apple.md) - Published: Oct 15, 2026 - Type: Benchmark Tool-augmented LLM agents must discover relevant APIs from large registries before planning, but finite context windows prevent injecting all tool definitions into a single prompt. Two competing architectures are Retrieve-then-Plan (RtP), which embeds the query and retrieves similar tools via nearest-neighbor search, and Plan-then-Retrieve (PtR), which first generates abstract tool stubs as retrieval queries. Evaluating on ToolBench with 46,984 API endpoints across 50 categories reveals a clear complexity-dependent crossover: RtP outperforms PtR on single-tool queries while PtR outperforms RtP on multi-tool queries. PtR’s stub-hallucination failure mode acts as a double-edged sword—on simple queries it abstracts away proper-noun signals, yet on complex queries it decomposes multi-faceted intent into functional sub-needs. PtR roughly doubles inference cost, which pays off on multi-tool queries but hurts on single-tool ones, motivating complexity-adaptive routing driven by RtP’s top-1 retrieval score. ### [Run fast, private agents on-device with LFM2.5-24B-A2B](https://yomu.fyi/post/run-fast-private-agents-on-device-with-lfm2-5-24b-a2b.md) - Company: [Hugging Face](https://yomu.fyi/company/hugging-face.md) - Author: Raahul Vignesh, Song Duong, Akhilesh Deepak Gotmare, Fernando Fernandes Neto, Jiahui Wang, Mayank Agarwal, nathan ranchin, Sergei Tilga, Sinoué GAD, Tim Seyde, Zetian Li, changyi yang, Simon Lee, Leonie Monigatti, Maxime Labonne, Karthik Macherla, Alfred Eng, Khoi Nguyen, Paul Pak, xx, Tarek Dakhran, Anthony Maltsev - Published: Oct 13, 2026 - Type: Announcement LFM2.5-24B-A2B is a 24-billion-parameter reasoning model with only 2 billion active parameters per token, designed for fast, private on-device agents. It uses a Mixture-of-Experts architecture with 40 layers, 64 experts per block, and top-4 routing, combined with gated short convolutions and GQA. The model produces an explicit chain of thought before answering and supports tool calling and multi-step tasks. Pre-training scaled to 33 trillion tokens, followed by mid-training and a five-stage post-training pipeline including supervised fine-tuning, multi-domain distillation, and agentic reinforcement learning. On benchmarks, it leads or matches larger models on tool use and instruction following, achieving 48.09 on LiveClaw and 96.75 on IFEval, while decoding 184 tokens per second on an Apple M5 Max. ### [How to Evaluate AI Agents From Tool Calls to Task Completion](https://yomu.fyi/post/how-to-evaluate-ai-agents-from-tool-calls-to-task-completion.md) - Company: [NVIDIA](https://yomu.fyi/company/nvidia-developer-blog.md) - Author: Elizabeth Goodman - Published: Sep 21, 2026 - Type: Explainer Standard LLM benchmarks are insufficient for evaluating AI agents because they score static outputs rather than multi-step tool-call sequences against a live environment. The post explains that agent evaluation requires two scoring layers: step-level (process scoring) checks each call’s validity and usefulness, while end-to-end (E2E) scoring checks only the final environment state. A trace logs every step and environment state, and metrics roll up through a hierarchy of benchmark, trial, task, turn, and step. Key metrics include task success rate, consistency, tool-call precision, argument accuracy, steps per success, and cost per success. The post argues that benchmarks are converging on tool use because deployed agents always use tools, and enterprise benchmarks should measure task completion against real APIs and policies.