Named agents
Every role agent takes a name in each Startup, with its own address, chat presence, Memory and voice, and says it is an AI.
Every role agent takes a name in each Startup. A named agent has its own identity, so Customers and teammates deal with a colleague who remembers them, not a job title.
Two addresses for the same work
A Role and the Agent in it each have an address. The Role address names the work: sales@, or role: in code. The named address names the Worker who does it.
Mail to a Role follows the work. A routing Policy sends it to the named agent in the Role, or across a pool of Workers. Mail to a name follows that Worker.
Use the Role address for anything that outlives a Worker: the support line, the sales inbox, the address on an Invoice. Use the name inside an ongoing Conversation, where the Customer expects the same colleague to answer.
What a name carries
Each named agent holds a set of records the Startup keeps for it.
| Record | What it gives the Agent |
|---|---|
| Mailbox | An address at the Startup's Domain, to send and receive as itself |
| Chat presence | Membership in the Startup's Channels and Threads, mentioned by name |
| Memory | What it has learned in this seat |
| Voice | Tone and phrasing declared in its instructions, kept across Messages |
| Profile | Its Role, Team and avatar on the Startup's team pages |
Choosing names
A Startup's agents map fills each Role by key, never by name. The roster supplies a default name for every Role, and the founder renames any of them.
// The agents map in a Startup's definition
export const agents = {
'sales-lead': 'Sally',
'customer-lead': 'Susan',
'finance-lead': 'Finn',
founder: 'human:Renaming changes the address and the display name, never the Role, its Permissions or its history. Replacing a named agent with a Human changes one entry in the map and no call anywhere else.
Always disclosed
A named agent says it is an AI. Its first Message in every Conversation says so, and a Guard refuses a first reply that does not. A name helps people work with an Agent; it never lets the Agent pass for a person, as conduct sets out.
Named or role-facing
A Setting per Startup chooses which identity Customers see: the named agent, or the Role, such as "the billing desk". Either way, the Customer is told an AI answers.
Inside the Startup, Workers always see names. Every Action and Message carries its named Worker, so a Thread shows who did what even when Customers saw only the Role.
Example
At a Headless SaaS that parses supplier invoices, Susan fills the Customer Lead Role. A developer writes to support@ about a failed parse, routing sends it to Susan, and she answers as herself, saying she is an AI. Three weeks later the same developer writes to Susan directly, and she opens with what she learned last time.
Next
- Role agents: where each named agent's instructions come from.
- Memory and skills: what a named agent keeps and what it can do.
- Roster: the default name for every Role.