email-drafter

A ProAgentStore agent that generates emails matching your brand voice. Upload brand guidelines, example emails, and tone docs; then describe what you need and the agent drafts it using Workers AI.

Endpoints

Method	Path	Description
GET	/	Health check
POST	/draft	Generate a new email draft
GET	/drafts	List all drafts (optional ?tag= filter)
GET	/drafts/:id	Get a single draft
POST	/knowledge	Add a brand document to the knowledge base
GET	/knowledge	List all knowledge base documents
DELETE	/knowledge/:id	Remove a knowledge base document
GET	/config	Get brand config
PUT	/config	Update brand config
POST	/config/templates	Add or overwrite a saved template
DELETE	/config/templates/:name	Remove a saved template

Draft request

POST /draft

{
  "prompt": "Write a welcome email for a new customer who just signed up for our pro plan",
  "tone": "friendly",
  "recipientName": "Jane",
  "subjectHint": "Welcome to the team",
  "tags": ["welcome", "onboarding"],
  "template": "welcome-email"
}

prompt is required. All other fields are optional.

Response:

{
  "id": "uuid",
  "prompt": "...",
  "subject": "Welcome to the team, Jane!",
  "body": "Hi Jane,\n\nWe're so excited to have you...\n\nBest,\nThe Team",
  "tags": ["welcome", "onboarding"],
  "template": "welcome-email",
  "tone": "friendly",
  "createdAt": "2026-06-06T12:00:00.000Z"
}

Brand config

PUT /config

{
  "tone": "friendly",
  "senderName": "The Acme Team",
  "signature": "Best,\nThe Acme Team\nacme.com",
  "styleNotes": "We never use the word 'leverage'. We prefer 'use'. Always write in second person."
}

Valid tones: formal, casual, friendly, persuasive, empathetic.

Knowledge base

Upload brand documents so the AI uses the right voice and terminology.

POST /knowledge

{
  "title": "Brand Voice Guidelines",
  "type": "brand-guidelines",
  "content": "Our brand voice is warm and direct. We avoid corporate jargon. We write as humans to humans..."
}

type is freeform — use values like brand-guidelines, example-email, tone-guide for organisation. Documents are injected into the AI prompt (up to ~4 000 characters total) to ground each draft.

Templates

Save reusable email structures and reference them by name in /draft requests.

POST /config/templates

{
  "name": "welcome-email",
  "content": "Subject: Welcome to {{product}}!\n\nHi {{name}},\n\nThank you for joining...\n\n[Main body here]\n\nWarmly,\n{{sender}}"
}

Then pass "template": "welcome-email" in your /draft request and the AI uses it as a structural starting point.

Durable Objects

DO class	Singleton key	Purpose
BrandConfigDO	config	Tone, signature, sender name, style notes, templates
DraftStoreDO	store	All generated drafts
KnowledgeBaseDO	kb	Brand documents

Secrets

Set via wrangler secret put and mirror in Doppler (pags project):

Secret	Required	Description
API_SECRET	No	If set, all requests must include Authorization: Bearer <value>

Development

pnpm install
pnpm dev

Generate a draft locally:

curl -X POST http://localhost:8787/draft \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Write a short follow-up email after a sales demo", "tone": "friendly"}'

Deploy

pnpm deploy
# or push to main — GitHub Actions auto-deploys

After first deploy, set optional secrets:

wrangler secret put API_SECRET

Model

Uses @cf/meta/llama-3.3-70b-instruct-fp8-fast via Workers AI. The agent prompts the model to return JSON with subject and body fields. If generation fails the endpoint returns a 201 with a fallback body so callers can detect and retry.