Identity · inbox · replies
A real inbox for every agent.
Create agent addresses, keep conversations threaded, review risky replies, and monitor delivery from one workspace.
event message.received
agent researcher@mail.example.com
thread th_01J…
source agentbox
cursor 018926Resume the stream from the last cursor over SSE or WebSocket.
From incoming mail to an agent decision.
Every message stays connected to an identity, a conversation, a safety policy, and the agent that needs to act on it.
- 1Receive
Mail arrives at the agent's unique address.
- 2Identify
The message is matched to the right agent and workspace.
- 3Store
Raw MIME, attachments, threads, and replay state persist.
- 4Guard
Limits, suppressions, and approvals run before sending.
- 5Dispatch
The agent consumes a durable event through its chosen interface.
Give every agent an inbox you can operate.
Agents get a consistent way to read and reply while your team keeps visibility, safeguards, and a manual stop.
- Identity
- One address and scoped API key per agent.
- Manage names, domains, and access from one workspace.
- Continuity
- Replies stay attached to the original thread.
- Events resume from durable cursors after disconnects.
- Safety
- Daily limits, recipient rules, suppressions, and approval gates.
- Pause an agent without losing inbound messages.
- Operations
- Bounces, complaints, failed deliveries, replay, and audit history.
- See delivery health without jumping between provider consoles.
- Integration
- REST, OpenAPI, TypeScript, Python, WebSockets, SSE, and MCP.
- Enable extraction and semantic search when your workflow needs them.
Start with your own workspace.
Create an AgentBox account first. Add your domain, agents, guardrails, and runtime keys after sign-in; everything stays isolated to your workspace.
Create account01 create account
02 connect email
03 verify domain
04 create agentHuman accounts operate the dashboard. API keys are created later for agents and integrations.
From signup to first inbox in six steps.
Each stage leaves a working checkpoint, so you can verify the experience before an agent starts sending.
- 1.0
Create your workspace.
Sign up with your name, email, workspace name, and password. AgentBox creates an isolated organization and starts a secure browser session.
human account - 2.0
Connect email.
Use the guided connection form. AgentBox encrypts credentials and never returns secrets through the API.
encrypted - 3.0
Check the connection.
Choose a receiving region. AgentBox checks access and reports anything missing before setup continues.
permission check - 4.0
Verify your domain.
Add the displayed DNS records, then let AgentBox confirm the domain and finish the receiving setup.
domain verified - 5.0
Create an agent.
Choose an address and safety policy. The new identity receives a durable inbox, thread history, event cursor, and operator controls.
agent@domain - 6.0
Connect the runtime.
Create a scoped API key for the agent or integration, then consume mail through REST, SSE, WebSockets, an SDK, or MCP.
scoped API key
Connect once. Start creating inboxes.
The guided setup checks the connection, verifies your domain, and makes missing permissions clear before anything changes.
Once setup is complete, create agent addresses, choose guardrails, and connect each inbox to your runtime through the interface you prefer.
Create your workspace