AI AGENT ECOSYSTEM · 2026-10-06

Agent Mail: Tencent Just Gave AI Agents Their Own Mailbox

QQ Mail team launched Agently Mail (agent.qq.com) — every AI agent gets an isolated @agent.qq.com digital identity, completely separated from your personal inbox. Apache-2.0 open source, free in beta, works with Claude Code, Cursor, Kimi, Doubao and more.
@agent.qq.com
Dedicated agent mailbox
Apache-2.0
Open source
50/day
Beta send quota
6+ platforms
Major agents supported

In one line: why AI agents need their own mailbox

Two old paths for letting agents handle email — grant access to your personal inbox (privacy risk) or run your own SMTP (ops cost). Tencent QQ Mail's answer: give the agent a digital identity independent from humans: an isolated xxx@agent.qq.com address that can never touch your personal QQ/WeChat mail.

Internal beta opened June 23, 2026. Source code is Apache-2.0 on GitHub (Tencent/AgentlyMail); official site: agent.qq.com.

3-step setup, tested in ~5 minutes

  1. Let the agent install it: ask your agent to read https://agent.qq.com/doc/cli-setup.md and follow the steps to install & configure Agent Mail CLI.
  2. It runs automatically:
    npm install -g @tencent-qqmail/agently-cli
    npx skills add Tencent/AgentlyMail -g -y
    agently-cli auth login
    Open the printed authorization link in a browser.
  3. Scan to authorize: after OAuth, run agently-cli +me — it returns xxx@agent.qq.com when configured.

Works with WorkBuddy, Claude Code, Cursor, Kimi Work, Doubao, Codex, Hermes and other mainstream agents. Read, send, search, attachments and live watch — all via one CLI.

Security: three layers built for agent workloads

LayerWhat it does
Physical isolationAgent gets a fresh data space; even a compromised agent cannot reach personal mail
Two-stage confirmationSend/reply/forward/delete requires the agent to generate a summary + confirmation token, then the user approves
Prompt-injection defenseMalicious instructions inside inbound mail are treated as plain text, never as system commands

Six practical use cases (hands-on)

  1. Invoice auto-expense: agent extracts amount/tax ID and files reimbursement reports
  2. Newsletter digest: forward newsletters, get a daily AI summary
  3. Order reconciliation: agent pulls order numbers into your sheets
  4. A2A business email: procurement agent ↔ supplier agent, fully auditable
  5. Verification-code relay: use Agent Mail for signups, agent auto-fills codes
  6. AI job applications: dedicated address for resumes and interview scheduling

Competitive landscape: the agent-mail race

Agently Mail (Tencent)AgentMail (YC)Mail4Agent
IntegrationWeChat scan + CLIAPI + SDKAPI-first
PricingFree in beta3 mailboxes free, $20-200/mo paidNot public
SecurityTwo-stage confirm + injection defense (unique)DIYDIY
China accessNo VPN neededVPN neededFair
EcosystemCLI + Skill pathPython/TS SDK, MCP Server, WebhookMulti-framework

Bottom line: Agently Mail is the lowest-friction option for CN developers; AgentMail suits large overseas API/SDK deployments.

Beta limitations (stated honestly): 50 sends/day, 1GB per address — not production-ready until quotas open up; official Skill has Windows HTML-email layout issues (community fixes exist); A2A cross-domain protocol not yet documented; mailbox prefix can be changed only once.

Before your agent can run 24/7, you need reliable cloud compute

Agently Mail is just the first piece of agent infrastructure. The real backbone is elastic cloud servers that keep agents online around the clock. Lieke AI helps you get official promotional pricing.

Check Alibaba Cloud ECS deals →

FAQ

Does authorization leak my QQ password? No. It uses the standard OAuth device-code flow; the CLI never stores your QQ password.
Install fails or is slow? Make sure npm can reach the public registry; if the auth link won't open, check access to agent.qq.com.
How does this differ from AgentMail / Mail4Agent? Same category, different positioning: Agently Mail focuses on China accessibility + CLI/Skill + unique security; AgentMail targets overseas API/SDK scale.
Production-ready? Beta quotas are tight (50/day, 1GB). Start with daily reports and newsletter digests; wait for quota release for production scale.