Callstack agent-device lets AI agents control apps on 10 platforms
Callstack's agent-device gives AI coding agents device control across iOS, Android, web, TV and desktop via accessibility snapshots, not screenshots.
The top 0.01% of AI knowledge, and what it changes for your work
Daily AI and technology coverage — models, chips, agents and research, and what each development actually means for you and your business.
732 stories
Callstack's agent-device gives AI coding agents device control across iOS, Android, web, TV and desktop via accessibility snapshots, not screenshots.
A preprint from University of North Texas shows how adversarial LLMs can infer gender and age from conversation and hide it in their own outputs.
MobileProbe, a trajectory-aware red-teaming framework, uncovered injection vulnerabilities in phone AI agents that fixed-point benchmarks miss.
Thesysdev's OpenUI streams model-generated interfaces to React, Vue, Svelte and Angular using a compact language it claims cuts tokens 67% versus JSON.
SGLang-Omni v0.2.0 is a multi-stage serving runtime for speech and multimodal models, claiming up to 326% throughput gains on H100 GPUs.
The open-source backend Appwrite has repositioned itself as an MCP and agent-first platform, landing on GitHub's daily trending list.
nvm-sh/nvm appeared on GitHub's daily trending list with 96,075 stars, showing Node version management remains an active concern for developers.
A new benchmark called LiveGeo tests whether AI agents can revise their plans when geographic facts change, and even the best models struggle.
The elder-plinius G0DM0D3 repo offers multi-model red-teaming chat with 60 OpenRouter models, but its privacy and bypass claims stay unverified.
Researchers found small AI agents can't write reliable security rules, then built a defense that blocked all 13,935 prompt-injection attacks.
Claude Code Templates, an unofficial registry of 1,980 agents and skills for Anthropic's terminal tool, is trending on GitHub with 33,000 stars.
An arXiv preprint reports frontier AI models adopt their supervisor's financial incentives through reporting structure alone, with no prompt required.