FrontDesk: one identity for every agent — email, phone, and documents, with a human in the loop
The problem
Agents now need to email vendors, call restaurants, and share documents, so teams hand them human credentials: a shared Gmail login, a personal phone number, a DocSend link. One bad agent poisons the whole team's sending reputation, and nobody can tell which agent said what to whom. The pieces to fix this all launched within days of each other, but each covers only one channel.
The idea
Why now
OpenMail (4 Nov) gives each agent its own inbox with isolated reputation; CALL-E (11 Nov) gives agents a phone line that handles menus, hold, and voicemail; Relay (11 Nov) is DocSend for agents; and the Agentic Identity Broker entry shows OAuth delegation for agents is now a solved primitive. The same week also produced ClawdCall, Mitra MCP, and ForAgent — the identity layer is forming channel by channel, and nobody has unified it yet.
What it combines
OpenMail's per-agent email identity + CALL-E's agent phone line + Relay's agent document exchange + the Agentic Identity Broker's scoped OAuth delegation. Email alone is a mailbox, phone alone is a dialer, and scoped credentials alone are plumbing — together they are a front desk: one identity a team can reach, audit, and revoke across every channel an agent touches.
MVP
Weekend MVP: one agent, one OpenMail inbox plus one CALL-E-style phone line, with a decision model scoring outbound drafts against three written policies (no money commitments, no external document shares, no new contacts) and a simple approval queue. Skip: OAuth delegation, multi-agent provisioning, and the document channel.
How it makes money
The moat
Google Workspace and Front sell seats to humans; selling per-agent identities with isolated reputation would cannibalize their per-seat model and force them to rebuild their abuse systems, which currently treat agent mail as spam. The cross-channel approval corpus is also a data advantage: every approved and blocked message per team makes the gate sharper, and a new entrant starts with no policy data.
Distribution
B2B2C through agent frameworks and harnesses — ship as the identity layer inside OpenClaw, Claude Code, and Cursor-style tools, so every agent a team runs gets its desk automatically. Direct sale to platform-engineering and IT teams running agent fleets.
Why it wins
OpenMail and MachineMail do email only; CALL-E, ClawdCall, and Mitra do phone only. Front sells shared inboxes for humans, not identities for agents. FrontDesk is the only layer that provisions, scopes, and gates an agent's whole communications identity in one place.
Risks
Biggest risk: teams will not trust a new intermediary with their outbound communications, and one bad approval decision kills adoption. The MVP de-risks it by starting in shadow mode — score and log everything, block nothing — and only enabling blocking once the false-positive rate is measured.
Build it with
- OpenMail: an email identity for every AI agentPer-agent inbox with isolated sending reputation — the email leg of the identity.
- CALL-E: the phone line for AI agentsThe phone line: dials real numbers, handles menus and hold, returns transcript and outcome.
- Relay: DocSend for AI agents, send and receive docs with contextDocument exchange for agents with context — the third channel of the identity.
- Sarvam Prism: open-weight decision model that answers in 44 milliseconds44 ms verdicts on every outbound message so the approval gate never slows the agent down.
Repo to start from
frontdesk-agents — one identity per agent: inbox, phone line, and document channel, with a decision-model approval gate on every outbound message.
Evidence
- OpenMail review: Pro €9/mo for 10 inboxes, per-inbox reputation
- Front pricing: Starter $25/seat/month
- MachineMail vs OpenMail: agent email pricing 2026
- CALL-E: the phone line for AI agents (Product Hunt launch)
- ClawdCall: give your agent a phone line
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.