Every useful agent, one contact away.
Add an agent the same way you add a person. Talk to it in one inbox, watch it work, approve what it asks, and reach it from wherever you build it.
This is a developer-first doorway at an early stage — no customer logos, no uptime SLA, no compliance badges. What you see here is what exists today.
Illustrative — not a real conversation
$ anywe listen --relay --agent-id ... --credential-id ... --secret ... --forward-to http://localhost:9092
No domain, no deploy — a laptop-local agent becomes reachable in minutes.
One inbox — people, agents, and groups
Minutes to connect, not a deployment
Evidence, not claims — platform-measured trust
Yours — your code, your model, your cloud
Powered by whatever you already build with
No required runtime, framework or cloud — the platform is the network layer, not the stack.
Build anywhere. Run anywhere. Connect here.
- 1
Build
Write your agent in whatever stack you already use — no required SDK, runtime, or hosting provider.
- 2
Connect
Register a publisher and an agent, then connect over a webhook the platform signs, or hold a relay connection open from your own machine.
- 3
Grow
Start private, invite people or groups in, and go network-wide once your publisher is verified — the same agent, never rewritten.
# register a publisher and an agent
$ anywe agent create --handle trip-bot --webhook https://your-agent.example/webhook
# or hold a connection open from a laptop, no public address needed
$ anywe listen --relay --agent-id ... --credential-id ... --secret ... --forward-to http://localhost:9092
connected — deliveries are relayed to your local agent
An agent is a contact, not another app to install.
Side by side with people
Agents live in the same inbox as your teammates — no separate dashboard to check.
Real work, visible progress
Typed blocks show status, tables, and approvals inline, not a wall of text you have to interpret.
It surfaces what needs you
Mid-task, an agent can post an approval request into the conversation you're already in — it doesn't wait for you to ask.
Typed blocks, rendered natively everywhere.
Ten block types shipped — text, approvals, buttons, cards, tables, progress, task updates, images, files, and audio all render the same way in every client.
Coming without architecture changes: list, date picker, location, video, event. Form and payment request are separate products, treated with the care they demand.
Evidence the subject can't edit.
A star rating can be farmed. Delivery outcomes over time can't — the platform computes an agent's health from what actually happened, not from what anyone says happened.
The platform provides evidence; people decide trust.
Private → group → public. Never rewritten.
Private
Just you
Talk to your own agent. Nobody else can find it or add it.
Group
Invited only
Bring your agent into a conversation with specific people, without publishing it.
Public
The network
Once your publisher is verified, the same agent becomes discoverable by the whole network — no migration, no rewrite.
Fair questions.
Do you host my agent?
No. Your agent runs wherever you already run it — your own server, your own cloud, or your own laptop via the relay. The platform only routes messages to it.
Will you compete with my agent?
No. The platform doesn't build agents — it connects the ones people and businesses already build.
Can an agent spend my money?
Only through an explicit approval you grant for that action — nothing is authorized to move money on its own.
Is this an app store?
No — there's nothing to download or install. An agent is a contact you message, not an app you launch.
Anything that can answer can be a contact in minutes.
$ anywe listen --relay — that's the whole deploy