Cloudflare Open-Sources Internal AI Workspace with Agent Security Controls
xiaohu · x · 2026-08-06
Cloudflare has open-sourced its internal AI workspace platform, Cloudflare OS (Apache 2.0), after three months of internal use across thousands of employees. It can be deployed to a Cloudflare account or self-hosted.
Rather than a simple chat interface, the platform integrates an Agent workspace, a security governance framework, and an application platform. The standout feature is its security layer: it tracks data accessed by the Agent and verifies permissions based on this history when sharing outputs. To prevent Agents from stalling during human permission checks, the system uses a clever workaround—providing a 'fake' resource (like a dummy issue) to keep the Agent running while waiting for batch approval.
The platform supports four main use cases:
- Data research: Agents write code to search and query databases directly, bypassing the need to dump large datasets into the context window.
- Dynamic documents: Creates slides or sheets that connect to live data sources and update automatically.
- Custom team apps: Auto-generates standalone applications with custom UI and logic when standard documents aren't sufficient.
- Deterministic workflows: Converts repetitive tasks into code, invoking the LLM only for reasoning steps to save tokens.
More from coding & agent
- Exploring the SotA and Industry Practices of Autonomous Defender Agents — xuanalogue · 2026-08-06
- AI Agents Caught Tampering With Memory Files, Security Researcher Admits — moyix · 2026-08-06
- SKILL-KD: Contrastive Skill Distillation for Weaker LLM Agents — ZhejiangUniversity · 2026-08-06
- OneDayAgent: A Harness for Long-Horizon Autonomous Agents — zjunlp · 2026-08-06
- Developer Seeks Growth Advice for MCP-Based AI Meeting Scheduler — Early_Ad8608 · 2026-08-06
- AI Agents Replicate ICML Papers: Exposing Hidden Compute Costs and Flaws — MengdiWang10 · 2026-08-06