Three open-source models sketch a desk-scale local agent loop: brain, voice, and eyes
lmoroney · x · 2026-10-27
Laurence Moroney combines three open releases from Oct 22-24 into a cloud-free local agent stack: Nemotron 3.5 Bolt E4B as the brain (NVIDIA's 8B-param/4B-effective model post-trained from Gemma-4-E4B-it, 4GB at Q4, 64K context, 15.5 tok/s on Jetson Orin Nano 8GB), Phonon for local speech via Ollama 0.40.11 on macOS with an OpenAI-compatible TTS endpoint, and WeVisOne-VL for grounding camera crops with boxes and masks. He details how to wire them as specialists and flags vendor claim gaps (Ollama lists 128K context vs NVIDIA's 64K).
Related event: Three Open-Source Models Combine Into a Desktop-Ready Local Agent(2 posts)→
More from coding & agent
- Anthropic: Single agents with skills often beat subagent designs in enterprise deployments — addyosmani · 2026-10-27
- Claude Code's idle compaction annoys users — here's the setting to turn it off — JasonBotterill · 2026-10-27
- Creator builds CCTV-style video intro with Claude Code, still stuck on AI dubbing — xiaohu · 2026-10-27
- Open-Source Agent Project GEAR Trends on Hugging Face — WizardTY · 2026-10-27
- Skill∞: Recursive Self-Improvement Where Agents Optimize Their Own Optimization Skills — NationalUniversityofSingapore · 2026-10-27
- Agent identity by name is untrustworthy: 5 runtime principles for multi-agent credential safety — karc16 · 2026-10-27