Stay in the Loop

On-device phone agents: what open mobile agents actually bring (after OpenMobile-2)

OpenMobile-2 opens up mobile agent training environments and recipes, and LFM3-1.7B shows a 1.7GB agent already runs on your phone: numbers, minimum requirements and honest limits.

ML Drift: Google's open-source on-device GPU engine, and when it beats the generalists

The engine that runs models inside Chrome and YouTube is now open source: architecture, benchmark numbers, and an honest comparison with Ollama and llama.cpp.

Nemotron 3.5 Bolt: NVIDIA's 74B MoE with ~9B active — what changes on your desktop

74B parameters, 9B active: NVIDIA's MoE for local agentic work fits 24GB of VRAM via IQ2 quantization. Real VRAM, declared benchmarks, and honest limits.

Local text-to-SQL: query your business database without sending data to the cloud

A local model, the schema to the model, execution on your database: text-to-SQL in-house, ARCS benchmark numbers, and where it really breaks.

MOSS-Image: OpenMOSS-Team's ComfyUI ripack, workflows and real requirements

MOSS-Image: OpenMOSS-Team's ComfyUI ripack, ready in two folders. Real-world VRAM computed per weights combination, from ~11GB to ~31GB, plus the four included workflows.

Mistral Large 4: a 1-trillion-parameter MoE, 49B active, released open-weight

Co-Design Bench: 273 tasks from real user requests to evaluate multimodal coding agents

Gartner: 9 in 10 agentic workflow steps are routine, and open-weight models handle them at a twentieth of the cost

LiquidAI LFM3-1.7B: a 1GB-at-4-bit edge agent model with 256K context and 16 languages

Cost-Aware Abstention: on 47,837 IT tickets, a thresholded TF-IDF classifier beats LLMs on cost