#3232nd of 35 in Agent platforms
cloudflare-os
Agent chat workspace where AI builds sandboxed per-user apps with approval-gated access
- Stars
- 11.5k
- License
- Apache-2.0
- Last commit
- Oct 2026
- Language
- TypeScript
3 on 14 Apr 202611,517 on 20 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Cloudflare is hiring: 440 open roles
Overview
Cloudflare OS is an internal Cloudflare productivity environment, now open source, built on Workers, Durable Objects and Dynamic Workers. Users chat with a Code Mode agent that does tasks directly or builds small private apps called gadgets, each in its own sandbox. External services are reached through Gatekeepers, which log actions and queue side-effecting ones for later human approval. The README labels it early access.
Who it is for: Teams building an internal, security-gated AI workspace on Cloudflare Workers
Strengths
- Gadgets run in Dynamic Worker sandboxes with internet disabled and iframe-isolated clients
- Gatekeepers simulate side effects; users approve or reject queued actions in bulk later
- Capability-based access: agents and gadgets get no resources until explicitly introduced
- Runs locally with `pnpm run-local` on wrangler and workerd; works with many LLM providers
Weaknesses
- README calls it early access with many rough edges; this is a version 2 rewrite
- Self-hosting on plain workerd is marked COMING SOON; local mode is not for production
- Depends on Workers features such as Durable Objects, Facets and Dynamic Workers
- Each Gatekeeper needs its own OAuth credentials; outside contributions largely not accepted
What it needs
- no GPU
- Needs pnpm, wrangler, workerd, Cloudflare account (for hosted deploy), LLM provider
- Models: Many major LLM providers via pi-agent-core, self-hosted models
- port 8787
Also in Agent platforms
See all 35| Rank | Project | Score |
|---|---|---|
| 1 | LangflowVisual builder for AI agents and workflows, exposed as API or MCP server | 81 out of 100 |
| 2 | daily_stock_analysisDaily LLM stock analysis for A-share, HK, US and other markets, pushed to chat apps | 80 out of 100 |
| 3 | deer-flowAgent harness that runs sub-agents, memory and sandboxed skills for long tasks | 77 out of 100 |
| #4 | vibe-tradingFinance research agent for backtests, market data and broker connectors | 77 out of 100 |
| #5 | MulticaIssue board where AI coding agents take assignments like teammates | 76 out of 100 |