Skip to main content

Layer 3: Full-stack primitives

The app around the agent: compute, data, files, durable pipelines, schedules.

In this layer: Workers, D1, R2, KV, Workflows, Queues (plus Cron).

Workers​

What it is: Serverless compute on every Cloudflare location. Your agent, your API, your Custom Domain.

Use it when: Always: your Worker serves the chat UI, POST /api/chat and /mcp.

// wrangler.jsonc: your team card gives the account id and the hostname
"account_id": "<TEAM_ACCOUNT_ID>",
"routes": [{ "pattern": "<TEAM_HOSTNAME>", "custom_domain": true }]

Basics: Workers

Advanced: Custom Domains

D1​

What it is: Serverless SQLite. Tables, joins, migrations.

Use it when: An audit log of what the agent proposed and what a human approved.

// npx wrangler d1 create agent-db
await env.DB.prepare('INSERT INTO actions (id, proposal, approved_by) VALUES (?, ?, ?)')
  .bind(id, proposal, approver).run();

Basics: D1

Advanced: Migrations

R2​

What it is: Object storage with no egress fees.

Use it when: Storing Browser Run screenshots as evidence, or reports the agent generates.

await env.BUCKET.put(`evidence/${id}.png`, png, { httpMetadata: { contentType: 'image/png' } });

Basics: R2

Advanced: Workers API

KV​

What it is: Global key-value store for read-heavy config and caches.

Use it when: Feature flags or a known-good baseline your watcher compares against.

await env.KV.put('baseline:se', JSON.stringify(vitals), { expirationTtl: 3600 });

Basics: KV

Advanced: KV API

Workflows​

What it is: Durable multi-step execution with retries, sleeps and a pause for a human.

Use it when: Publish, purge both zones, verify, and wait for approval before the risky step.

const plan = await step.do('plan purge', async () => planPurge(event.payload));
const ok = await step.waitForEvent<{ approved: boolean }>(
  'await approval', { type: 'purge-approval', timeout: '1 hour' });
if (ok.payload.approved) await step.do('purge', () => purge(plan));

Basics: Workflows

Advanced: Events and parameters

Queues (plus Cron)​

What it is: Buffered, retried background work. Cron Triggers run a Worker on a schedule.

Use it when: Fan out a scan of every market page, or run a watcher every 15 minutes.

await env.QUEUE.send({ market: 'se', url });
// wrangler.jsonc: "triggers": { "crons": ["*/15 * * * *"] }

Basics: Queues · Cron Triggers

Advanced: Agents SDK schedules