AgenticEmail examples

Working AI email agents you can clone and run in minutes. Each one gives an agent its own AgenticEmail inbox - it sends, receives, and runs full email threads on its own.

Every example ships in Python and TypeScript, mirrors a template at agenticemail.dev/build, and is self-contained: clone the folder, set two environment variables, and run.

Examples

Productivity

Example	What it does
scheduling-agent	Email it to book time - it knows your rules and offers slots.
inbox-zero-agent	Drafts replies while you sleep - you review and send.

Sales & outbound

Example	What it does
gtm-agent	Personalised outreach, reply classification, warm handoff.

Support

Example	What it does
support-agent	Triage, respond, escalate, follow up, close.

Recruiting

Example	What it does
ai-recruiter	Screen inbound applicants, coordinate interviews, keep candidates warm.

Finance

Example	What it does
invoice-processor	Claude reads PDF invoices, matches, and routes them by amount.
receipt-parser	Claude vision reads receipt photos and PDFs into clean expense rows.

Verification

Example	What it does
signup-verification-agent	Give a browser agent an inbox to catch signup links and 2FA codes.

Personal

Example	What it does
newsletter-digest	A daily digest of everything that hit your inbox.

Quickstart

1. Get an AgenticEmail API key at app.agenticemail.dev/keys (free tier, no credit card).

2. Get an Anthropic API key at console.anthropic.com.

3. Pick an example and a language:

   Python (3.10+)

   cd scheduling-agent/python
   pip install -r requirements.txt
   export AGENTICEMAIL_API_KEY="am_..."
   export ANTHROPIC_API_KEY="sk-ant-..."
   python agent.py

   TypeScript (Node 18+)

   cd scheduling-agent/typescript
   npm install
   export AGENTICEMAIL_API_KEY="am_..."
   export ANTHROPIC_API_KEY="sk-ant-..."
   npm start

4. Email the address the agent prints on startup, and watch it work.

Each example's own README lists the config (timezone, escalation address, approval limits, and so on) you can tune.

How they work

Every agent uses two SDKs:

• agenticemail (npm) - the inbox: create it, poll for new mail, send and reply.
• anthropic - Claude reads each message under the agent's system prompt and writes the response.

The loop is the same shape across examples: create an inbox, list new inbound messages, hand each one to Claude with the system prompt, and send back what Claude writes. The prompt lives in its own module (prompt.py / src/prompt.ts) so it is easy to edit.

Prefer to skip the code?

Connect AgenticEmail to Claude Code, Cursor, Windsurf, or Cline over MCP and your existing agent gets an inbox with zero glue code - see agenticemail.dev/build.

Verify it works

Every example uses the same inbox path (create, send, receive, reply). One smoke test exercises it end-to-end against the live API, plus the Claude text-block extraction the agents rely on:

pip install agenticemail anthropic
export AGENTICEMAIL_API_KEY="am_..."
export ANTHROPIC_API_KEY="sk-ant-..."   # optional
export INBOX_DOMAIN="yourdomain.com"    # optional, if your org has >1 domain
python scripts/smoke_test.py

It reuses a single inbox (fixed client_id) so it never accumulates inboxes. CI (.github/workflows/smoke.yml) runs it on every push and weekly - add AGENTICEMAIL_API_KEY (and optionally ANTHROPIC_API_KEY, INBOX_DOMAIN) as repo secrets to enable it.

Multiple verified domains on your org? Set INBOX_DOMAIN - create() needs a domain when there is no default.

Learn more

• Docs: agenticemail.dev/docs
• API reference: api.agenticemail.dev/docs

License

Apache-2.0