Postfleet field notes

Email systems agents can trust.

Security patterns, typed extraction, and working integrations for engineers who let software read the inbox.

Latest reports

Start with the trust boundary

SecurityEmail prompt injection: how to secure an AI agent that reads emailTreat incoming email as hostile data, screen it before model context, restrict the tools it can reach, and require approval for consequential actions.July 12, 20267 min read

Engineering reliabilityBest email APIs for AI agents in 2026Compare Postfleet, AgentMail, Nylas, and Resend by inbox model, event delivery, agent controls, and the engineering work each leaves to you.July 17, 20268 min read

Structured extractionEmail to JSON with AI: a JSON Schema-first guideTurn email and attachments into typed JSON by separating MIME parsing from semantic extraction, validating output, and preserving honest failure states.July 12, 20266 min read

TutorialsGive an OpenAI agent a secure email inbox with MCPConnect the OpenAI Agents SDK to a Postfleet mailbox over Streamable HTTP, then narrow the key, tools, and approval boundary for production use.July 12, 20266 min read

Proof before promises.

These articles use public standards, upstream documentation, and Postfleet's shipped behavior. When a defense has limits, the limits stay in the article. When code can fail, the failure shape is part of the example.