Cloud.ru

Latest AI news, models and releases from Cloud.ru. ['Evolution Foundation Models', 'GigaChat 3.5 Ultra']

Agents 🇷🇺

AI agent failures: why dashboards are green but agents lie for days

Classic monitoring doesn't catch AI agent failures because agents degrade gradually, not crash. To detect issues, teams must trace reasoning cycles, measure agentic quality metrics, use LLM-as-judge evaluation, version prompts, and write telemetry to two destinations from day one.

Cloud.ru

Habr — хаб ИИ29.07 · 12:03

Open Source 🇷🇺

Cloud.ru Open-Sources Guardrails Filter for Data Protection in LLM Workflows

Cloud.ru has released the source code of its Guardrails Filter, a transparent reverse proxy that removes sensitive data from requests to LLMs and restores it in responses. The tool includes ~260 built-in rules for detecting and masking personal data, API keys, and more, with support for streaming and tool calls. It is available as a standalone service or as an Envoy ext_proc sidecar.

Cloud.ru

Habr — хаб ИИ24.07 · 06:01