Your Inbox, safely shared with agents.
Let AI agents draft and reply to your email without giving them your whole inbox. Agents see only the messages you allow, and you see everything they do.
Share your inbox in four steps.
Write plain-English rules for what each agent can read, draft and send. No agent gets your whole inbox.
Connect Gmail to Clam
Clam classifies your email into categories your rules can use, like travel or receipts. It never stores the email content.
Create a Shell for each agent
A Shell is one agent's slice of your inbox, with access rules you write in plain English. A travel agent's Shell might say: read my flight and hotel emails, and draft replies but never send them.
Install your agent connector
One click connects your agent to Clam's MCP server. It gets the email tools it already knows, but only for the emails you allowed.
Put your agents to work
Your agents triage new mail, sum up long threads, draft replies and archive what's done, all inside their Shell. A deterministic policy check lets each one do only what your rules allow, and Clam logs every move.
One inbox, a Shell for every job.
Each Shell starts as plain-English rules. Clam compiles them into Cedar policies and checks every request to your inbox against them.
Your rules, enforced on every call.
Allowed actions pass through and everything else is blocked. Clam logs every read, draft and blocked attempt, with the exact policy rule that decided it.
Least privilege
An agent gets only what its Shell's rules allow.
No email stored
Clam proxies requests. Gmail stays the source of truth.
Full audit trail
Every action an agent takes or attempts, with the decision and the rule behind it.
Live decision log

ChatGPT search "invoice 1043"
Mar 28, 2024, 2:30 PM

Claude read verification codes
Mar 28, 2024, 2:30 PM

Gemini send draft → vendor@newco.io
Mar 28, 2024, 2:30 PM
Hermes delete conversation
Mar 28, 2024, 2:30 PM
Muse read calendar invite
Mar 28, 2024, 2:30 PM
Grok read label:Payroll
Mar 28, 2024, 2:30 PM
Muse search from:vendor.com
Mar 28, 2024, 2:30 PM
Secure at every layer.
Clam stores no email content. Your Google tokens stay with Clerk, our authentication platform, never in our database. Every agent request passes a deterministic Cedar policy check before it reaches Gmail, and our models can label mail, never act on it. Traffic uses TLS 1.2 and 1.3, data is encrypted at rest with AES-256, and our services talk over a private WireGuard network.Read our security manifesto.
- CASA AL1ValidatedOct 2026 · TAC Security
SOC 1In auditExpected Nov 2026
SOC 2In auditExpected Dec 2026
CASA AL1 assessed by TAC Security · SOC reports expected Q4 2026
Where Clam is going
Clam opens as a capped beta on Gmail in mid November. These are the milestones after it.
MVP
Capped Beta Launch
Early adopters get access by invitation and tell us what to fix.
GA
General Availability and Outlook support
Clam opens to everyone. Connect an Outlook mailbox to Clam as well as Gmail.
V2.0
Clam for Business
Clam for Business brings company-wide rules, approvals and audit trails to Google Workspace and Microsoft 365.
MVP
Mid November 2026
Capped Beta Launch
Early adopters get access by invitation and tell us what to fix.
GA
Mid December 2026
General Availability and Outlook support
Clam opens to everyone. Connect an Outlook mailbox to Clam as well as Gmail.
V2.0
Q1 2027
Clam for Business
Clam for Business brings company-wide rules, approvals and audit trails to Google Workspace and Microsoft 365.
Frequently asked questions
No. Clam proxies requests. Gmail stays the source of truth, so there is no second copy of your mailbox to worry about.
Yes. Data is encrypted in transit and at rest, and every agent request is checked against your rules before anything is returned.
Only what that Shell's rules allow. The agent uses normal email tools (read, search, draft, send) and Clam blocks everything else.
In plain English. Describe what the agent may read, draft, or send, and Clam turns that into enforceable policy on every request.
Any MCP-compatible agent, including Claude, ChatGPT, Codex, OpenClaw, Gemini, and Grok. You point the agent at the Shell's hosted MCP endpoint.
Yes. Pause or delete a Shell and that agent loses access immediately, without touching your other Shells or your Gmail connection.
Yes. Every action and decision is in the audit log: what the agent did, what it tried, and which rule allowed or blocked it.
Still have questions?
Email us at support@useclam.ai

Give your agents an inbox you control
Share your inbox with your first agent in minutes and see every action it takes.
Get started with





