How they use Nebius Token Factory
Rasa uses Nebius Token Factory in voice-agent demonstrations and educational code. Developer relations engineer Rod Rivera described a live banking voice-agent demonstration built with Rasa and Speechmatics. Switching to Gemma3 27B through Token Factory improved speed and provided enough conversation capacity to make the demonstration useful.
Rasa’s companion repository for “Why (Voice) AI Agents Fail” uses Nebius inference across four runnable demonstrations covering turn-taking, backchannels and interruptions, concurrent state, and premature goodbyes. Token Factory supplies small language models for classifiers and agent loops through the OpenAI-compatible Python SDK. The repository names Meta-Llama-3.1-8B-Instruct as its default model and reports sub-second inference with fast model variants. Separately, Rasa reports co-hosting an AI-agent engineering webinar with Nebius Academy.
Sources
3 sourcesWhy (Voice) AI Agents Fail companion code
19 Oct 2026 · GitHub · Web page
Rasa’s voice-agent course repository describes using Nebius Token Factory for inference, with an OpenAI-compatible SDK, a default Llama model and reported sub-second inference.
Open original sourceFrom Chatbots to Agents: Engineering Reliable AI Workflows
26 Feb 2026 · Rasa · Web page
Rasa reports co-hosting an AI-agent engineering webinar with Nebius Academy.
Open original sourceCompany context
rasa.comRasa develops a platform for enterprise product teams to build, test, deploy and monitor customer-facing voice and chat agents. Its tools combine language models with defined business logic, allowing teams to control how agents handle conversations and complete tasks. The platform includes a visual Studio workspace, developer tools, automated conversation evaluations and monitoring, with deployment on customers’ infrastructure. It serves organizations in banking, insurance, healthcare, retail and telecommunications, supporting agents across phone lines, web chat, Slack, WhatsApp and email.
Recent updates
1 updateRasa launches Mantle agent orchestrator
Rasa introduced Mantle, a new orchestrator for building agents with reusable skills and centralized governance. Teams can build in natural language, evaluate against simulated customers and improve agents using real conversations. Rasa said Mantle was available immediately.
Rasa