Live workshop Nov 3: CPU agent with GPU image and speech models as tools, Endpoint vs Job cost test
Al_Grigor · x · 2026-11-01
DataTalks.Club hosts a free live coding workshop on November 3, 16:30 CET, led by Alexey Grigorev and Mikhail Rozhkov (Nebius), on a "cheap agent, expensive models" architecture: the agent runs on a CPU, reads a brief and decides what to generate, while costly text-to-image and text-to-speech models live on GPUs and get called as tools only when needed.
Agenda (90 min):
- Architecture rationale: CPU for the agent, GPUs for the models
- Wrapping a text-to-image model on a Nebius Serverless Endpoint as a generateimage(spec) tool returning an Object Storage link
- Building the CPU agent as control plane, adding text-to-speech via MCP
- Running the same generation as a batch Job over 50 records and comparing latency and cost per asset against the Endpoint
- Capstone preview: a Personalized Media Factory, plus Q&A
Requirements: a Nebius account with credits, uv installed, and the workshop repo. Attendees leave with a working media agent and a rule of thumb for choosing Endpoint vs Job.
Related event: Workshop: run agents on CPU, call GPU models as tools(3 posts)→
More from coding & agent
- Kaigen controversy: creator fires back at performance critics, shares playable code — gdechichi · 2026-11-01
- Claude-built local HTML page compares nine Mac terminal fonts side by side — SeanOliver · 2026-11-01
- A copyable error report template for debugging ComfyUI workflows — no3us · 2026-11-01
- Stageshow: a Claude skill for better faux UI stills, animations and bento grids — MxMnr · 2026-11-01
- Reddit user asks how to make Claude, ChatGPT and Antigravity collaborate on one game project — LastHmn · 2026-11-01
- Karpathy's nanochat hits 59K stars: a full LLM training lab in one repo — techNmak · 2026-11-01