Building a Read‑Only Cloudflare Worker AI Security Console

说明与用途

构建了一个只读的 Cloudflare Worker AI 安全控制台，聚合分析数据，提供自然语言查询、摘要和 Ray‑ID 调查，同时限制 AI 权限并执行访问控制。

问题

Security teams have data but lack a fast, readable operational view that keeps AI from gaining unsafe authority.

结果

The console delivers a read‑only, AI‑augmented security dashboard that lets analysts query data in natural language, view curated queries, receive daily digests, and investigate Ray IDs—all while preventing the AI from modifying Cloudflare resources.

工作流

• Create a read‑only Analytics API token with minimal permissions
• Bootstrap a Cloudflare Worker named sentinel‑cf
• Protect the Worker with Cloudflare Access (policy, MFA, short session)
• Add a Workers AI binding named AI
• Create and bind a KV namespace for latest digests
• Configure an AI Gateway for observability and rate limiting
• Add runtime variables and secrets (CF_ANALYTICS_TOKEN, APP_HOSTNAME, AI_GATEWAY_ID, ALLOWED_ZONE_TAGS, ENVIRONMENT, DEFAULT_DIGEST_RANGE, DIGEST_TITLE)
• Deploy the Worker code (sentinel‑cf‑worker.js)
• Test each endpoint (/, /query, /allowed‑queries, /digest, /ray, /healthz)
• Codify the configuration with Terraform for production
• Schedule daily digests via Cron Trigger and store in KV

证据

The model should not be the administrator.

来源

Workers AI owns: Intent classification when deterministic matching is not enough.

来源

The right production pattern is: AI recommends, humans approve, Terraform or approved change control applies.

来源