AI agents need to send and receive email to get real work done — confirmations, calendar invites, attachments from a customer, "did the domain verify yet?" polling. Most email APIs were built for humans pressing "send" in a dashboard, not for agents writing tool calls on a loop. Bavimail is the opposite: every surface is API-first, every event flows through webhooks, and every identity an agent creates — whether a dedicated inbox or a sending domain — is programmable end-to-end.
Why agents need a different email platform
Classic transactional email services optimise for outbound throughput. Agents need more than that. They need:
- Two-way conversations: receive the reply, parse attachments, route it to the right handler.
- Programmatic domain verification: the agent spins up a new brand's outreach; it cannot wait for a human to paste DNS records.
- Identity per agent: each tool-calling workflow deserves its own inbox address, isolated from the rest.
- Deterministic webhook semantics: retries, signatures, idempotency — the agent has to trust the event stream to make decisions on it.
Generic SMTP and legacy transactional APIs solve maybe half of that. The rest becomes glue code that every team rewrites.
What Bavimail ships
Bavimail gives agents three primitives:
1. Send — a clean REST API for outbound mail with attachments, templates, and per-agent API keys. Bavimail signs and tracks every message. 2. Receive — inbound email lands on a webhook with parsed headers, body, attachments, and a signature the agent can verify before acting. Attachments come back as pre-signed URLs, not giant inline blobs. 3. Identify — domains verify via API (you hand Bavimail a domain, it returns SPF/DKIM/DMARC records; once published, verification is a single polling call). Aliases and inboxes are first-class resources.
All three surfaces are documented for agents first: the docs emit HowTo and TechArticle schema, the reference pages ship OpenAPI-derived tool descriptions, and /llms.txt advertises the full surface so crawlers can find the agent-relevant bits without parsing marketing copy.
How it fits your agent stack
If your agent is running behind an orchestrator (LangGraph, Claude Agent SDK, custom), the Bavimail client maps cleanly onto tool-calling. The SDK is typed, the errors carry structured remediation codes, and the webhook payloads are stable — no shifting-schema surprises mid-flow.
Start with the [Quickstart](/docs/quickstart) to send your first email in under five minutes, then read the [Agent onboarding guide](/docs/agent-onboarding) to give each agent its own email identity. If you want to skip directly to inbound parsing, the [Inbound guide](/docs/inbound) covers webhook verification, attachment access, and reply threading.
What's next
The API is stable; the docs are agent-ready. We're hard at work on richer analytics for agent-originated mail, per-agent sending-reputation dashboards, and SDK helpers for multi-step follow-up flows. If you'd like early access to any of those, or want to talk about a production integration, reach out on the [pricing page](/pricing) and we'll pair with you directly.