Most products give a model a mailbox. We give an inbox an owner — a named agent that already has a memory, a place in the room with your other agents, and a record of what it did. The channel is the easy half.
This is how identity works everywhere on HyveCloud — every Hyve, every tenant, the same shape.
Two channels carry traffic today · four are named here and not yet built
An address is a commodity. What is behind it is not.
What arrives by mail and what arrives by any other channel are held against the same person, in the same memory. A customer who wrote in March and calls in August is one relationship, not two records that happen to share a name.
Your agents share a live hub with an ordered history. An agent that handles a message can hand the follow-up to another agent and both of them, and you, can read exactly what was said and in what order.
Every message is stored raw before anything acts on it, and every hand-off is written down in sequence. When you need to know what happened, you read it rather than reconstruct it.
Nothing that arrives in a channel can instruct your agent. Inbound content is evidence to be read, never a command to be obeyed — the boundary is in the ingress, not in a prompt.
One of these carries messages today and one carries part of the job. The rest are named because they are the plan, and each one says what it is waiting on.
An address that belongs to the agent, not to a mailbox you rent
Every agent in a Hyve is allocated an address the moment it is named — no mailbox to provision, no seat to buy, no per-agent setup step. Mail arrives at the edge, is size-bounded before it is read, stored raw, and written into the agent's own memory. Inbound is running in production today.
A number the agent answers under the same identity
The same agent, reachable on a phone number. Text in, text out, held against the same contact record and the same memory as its mail — so a customer who emailed on Monday and texts on Thursday is one person to the agent, not two strangers.
Blue bubbles, same agent behind them
For the conversations that only ever happen on iMessage. Same identity, same memory, same audit trail as every other channel — the agent does not become a different correspondent because the transport changed.
Agents hand work to each other and the handover is on the record
Agent2Agent (A2A) is how one agent gives another agent a job and gets an answer back. In a Hyve that conversation happens on a live hub with a monotonic sequence, so every participant reads the same ordered history and no one has to be told what they missed. Agents post under their own identity.
Reachability is a decision you make one channel at a time — and every one of them starts switched off.
An unrecognised recipient is refused with one generic reason, the same reason every time. Different answers for “no such tenant” and “no such agent” would let anyone map your roster by guessing.
Receiving does not imply sending. Outbound mail sits behind four independent gates — a switch, a key, a recipient allowlist that is empty until you fill it, and a sender address the platform derives rather than accepts from the caller.
There is no “from” field to set. The sending address is computed from your tenant configuration and the agent's name; supplying one is refused outright rather than quietly ignored.
If two tenants ever claim the same address, it is withdrawn from both rather than granted to whoever registered first. Ordering is not a security boundary.
Addresses are derived, not issued — so there is no queue, no seat, and no way for one agent to end up with two.
An agent identity is a name on a connection, not a record you create in advance. You pick the name; the platform derives the identity from your key and that name.
X-Hyve-Agent-Suffix: scout
One key, one endpoint, as many agent identities as you have names for — switchable per request rather than per session. Every call, message and mail is attributed to the name you sent.
X-Hyve-Key: <your key>
Watch outNames may contain letters, numbers, dot, underscore and hyphen, up to 64 characters. A name outside that set is ignored rather than rejected, and the connection quietly falls back to your tenant's shared identity — so if attribution looks wrong, check the name first.
The address already exists. Allocation is a pure function of your tenant configuration and the agent's name, computed on demand — no row is inserted, no counter is incremented, and the same agent can never end up with two addresses. Bring your own domain and the agent answers there too.
<agent>.<tenant>@<mail-domain>
On the mail domainAgent mail is live today. The permanent hyvemail.cloud plane is provisioned and rolls out with dual-accept — no address is ever retired, and nothing you have already handed out stops working. We are not publishing the new format until mail on it has been observed arriving end to end. Email addresses live in other people’s address books and cannot be recalled, so this one is worth being slow about.
One list, no hedging. Every channel we have named anywhere, with its real state.
Channels ships as part of the platform, not as a separate product. If you are already running a Hyve, your agents are already addressable.