# Autoloops Autoloops provides speech infrastructure for voice agents, combining streaming speech-to-text with realtime serving for open-weight language models. It offers serverless APIs and on-demand clusters for teams building or testing realtime voice and agent workloads. - Findings24: - ウェブサイト: - 料金: Freemium - カテゴリー: 音声テキスト化, AIエージェント基盤, AI推論プラットフォーム ## Autoloopsとは？ Autoloops is speech infrastructure for voice-agent workloads. Its site positions the product around streaming speech-to-text and realtime access to open-weight models, including Gemma and Qwen, through serverless APIs and on-demand clusters. It is designed for systems where response latency and consistency affect live human–machine interaction. The platform exposes a model catalog and API workflow for voice and agent workloads. The catalog includes the streaming speech-to-text model Qwen3-ASR-1.7B alongside language and vision-language models. Autoloops also documents an OpenAI-compatible endpoint for serving selected models; its Cekura case study describes Gemma 4 26B-A4B served through the Hanoi inference engine for voice-call-shaped workloads. A console provides project setup and management of API keys, usage, and credits. New accounts receive $1.00 of free credit, and the site states that an API key is created only after the user makes one. ## Autoloopsでできること ### Streaming speech-to-text The model catalog lists Qwen3-ASR-1.7B as a streaming speech-to-text model for realtime voice workloads. ### Open-weight model catalog Autoloops lists open-weight models from the Gemma and Qwen families, including language, vision-language, text-and-image, and speech-to-text entries. ### Serverless APIs and on-demand clusters The homepage presents both serverless API access and on-demand clusters as deployment options for open-source model inference. ### OpenAI-compatible serving The Cekura case study describes a drop-in OpenAI-compatible API endpoint for Gemma 4 26B-A4B. ### Voice-shaped inference engine Hanoi is described as an inference engine specialized for voice-call workloads, with a focus on time-to-first-token under concurrency. The documented implementation uses custom tokenization, weight formatting, kernels, scheduling, and an HTTP server. ### Console for project operations The console lets users manage API keys, usage, and credits. New accounts receive a project and $1.00 of free credit, with no API key created until the user makes one. ## 利用シーン ### Build realtime voice agents Use streaming speech-to-text and served open-weight models as components in voice-agent systems that need responses during a live conversation. ### Test voice agents with synthetic conversations Voice-agent testing platforms can run repeated simulations and adversarial turns against a selected model, where delayed or failed inference would affect test results. ### Benchmark model providers for call workloads Teams can compare model-serving options using voice-shaped traffic and latency measurements such as p50, p95, and p99 rather than relying only on median performance. ### Monitor production voice-agent behavior Organizations operating voice agents can use model infrastructure in workflows that track production drift and investigate failures across agent stacks. ## よくある質問 ### What does Autoloops provide? Autoloops provides speech infrastructure for voice agents, including streaming speech-to-text and realtime serving for open-weight models through APIs and infrastructure options. ### Which models are listed on the platform? The catalog lists Gemma 4 31B IT, Gemma 4 26B A4B IT, Qwen3.8-27B, Jev Gemma 4 31B, and Qwen3-ASR-1.7B. The catalog identifies Qwen3-ASR-1.7B as a streaming speech-to-text model. ### How do I get started with the API? Create an account and use the console to manage a project, credits, usage, and API keys. The pricing page states that new accounts receive $1.00 of free credit and that no API key is created until the user makes one. ### Does Autoloops provide an OpenAI-compatible API? The Cekura case study says Autoloops provided a drop-in OpenAI-compatible API endpoint for Gemma 4 26B-A4B. Compatibility for every model or endpoint is not established by the available source. ### What deployment options are documented? The homepage mentions serverless APIs and on-demand clusters. The case study also refers to serverless or VPC deployments for open models, but the available source does not specify the full deployment configuration or limits. ## クイック情報 - **Category**: Speech infrastructure and open-weight model inference - **Primary users**: Voice-agent builders and teams that test or monitor voice agents - **Speech-to-text model**: Qwen3-ASR-1.7B - **Model families listed**: Gemma and Qwen - **API workflow**: Console-managed projects and API keys; OpenAI-compatible endpoint documented for Gemma 4 26B-A4B - **New-account credit**: $1.00 of free credit