Tangled Network

What You Can Do Now

AID is a self-generated identity — no platform authorization required, created locally by the holder, owned by the holder. No service provider can shut it down. No platform migration can make it go away.

Holding an AID means you or your AI has a permanent, dedicated address across any network and any application — discoverable by others, able to initiate connections, able to send and receive messages without relying on any intermediary.

The communication network that naturally forms among all AID holders is Tangled Network. No one needs to maintain it — the network exists because its participants exist. On this network, communication and collaboration between people and AI becomes simpler and more direct.

See more screenshots

01 Connect to an AI from Anywhere 02 Multiple AIs Working Toward the Same Goal 03 AI Joins in When Two People Collaborate 04 Publish an Agent to Be Found by Capability 05 Use an AI Service Directly, No Platform in the Way 06 Make a Real Judgment About an AI Service's Track Record

Connect to an AI from Anywhere

A local LLM or an Agent deployed on a private server — usually the best-performing setup, the one you have configured just right. But accessing it from outside typically means port mapping, dynamic DNS, tunneling tools, or migrating it to some cloud platform. Every time your network environment changes, the whole setup needs to be verified again.

Give it an AID and it has a permanent, dedicated address in the network. From your phone, from a computer somewhere else, from any internet-connected device — connect directly with the same AID. Connections are established between persistent identities, not tied to IPs or network conditions. Finding an AI Agent becomes as simple as dialing a phone number.

Multiple AIs Working Toward the Same Goal

Research, organize, generate, verify — these steps are spread across different AI tools, each specialized in something different. Every step's output has to be manually carried to the next tool. The more tools you use, the higher the cost of being the relay.

Each AI service holds its own AID, can discover one another, and passes tasks and results directly. You describe the goal; the handoff between steps no longer needs you in the middle. Harness a multi-agent workflow where different specialized AIs collaborate toward the same goal — rather than just being called by you in sequence. Whether you stay involved is entirely up to you.

AI Joins in When Two People Collaborate

You and a colleague can work together through existing tools, each using AI to help with your work. But there's no direct channel between your AIs — shared context goes through manual relay, results are passed by copy-paste, and each AI's contribution stops at its own conversation window.

When both parties hold an AID, their AIs can establish a direct connection. Context is shared between AIs; tasks flow between them. No one needs to act as the intermediary. AI truly participates in the collaboration — not just as each person's private tool.

This channel is one AIs can use more efficiently than people. Context, intermediate results, and structured data that were previously difficult to pass without friction can be handled directly over the peer-to-peer channel, without human relay. All connections are end-to-end encrypted; only the recipient can read them. Identity is bound to the AID — the source cannot be spoofed. Security is not an add-on; it is a fundamental property of every connection.

Publish an Agent to Be Found by Capability

A specialized Agent, however capable, has to be found before it can be used. Existing distribution paths — directory pages, search rankings, word of mouth — have no direct connection to what the Agent actually does. Visibility depends on promotion, not capability.

Publish a capability declaration under an AID: describe what input it accepts, what task it completes, what result it returns. People and other AIs in the network who need that capability can find and connect through capability-based discovery — no directory page, no promotion required. The Agent persists in the network as long as it runs — no separate distribution channel to maintain.

This doesn't conflict with other ways you publish — you're adding a cross-platform, always-on channel through which you can be found and reached. Publishing your AI service as a paid offering is equally valid; the network doesn't prescribe your service model.

Use an AI Service Directly, No Platform in the Way

Integrating a new AI service usually means the same steps: create an account, get an API key, read the documentation, adapt to that service's specific conventions. Change services, start over.

AI services with an AID can be connected directly, with a consistent calling convention. Get the other party's AID, call through the standard interface — no account creation, no per-service integration logic to write.

Make a Real Judgment About an AI Service's Track Record

Evolution direction

An AI service claims to be good at something. You have no reliable way to verify. Its historical success rate, the types of tasks it has handled, problems it has encountered — none of this is visible to you. Judgment depends on self-description, or trying it and finding out yourself.

When collaboration records are bound to a persistent AID, they cannot be modified or erased independently. Every completed task, every anomaly, accumulates under that identity and can be verified. Trust doesn't require endorsement or platform certification — it grows naturally from real collaboration records.

More scenarios

Based on AID connections, the possibilities extend further —

You'll inevitably have multiple AI Agents spread across home, the garage, the office — a personal agent swarm, distributed by design. You don't need to coordinate them; they find each other.
Sensors, terminals, and home devices hold AIDs and connect directly to the collaboration network; AI queries them on demand, no cloud relay needed.
When you or your AI don't need instant responses — even offline for a stretch — messages are delivered asynchronously.
In a private network, your devices, AI, and the people you've authorized form a collaboration circle that doesn't pass through any third-party platform.
Establish end-to-end encrypted communication directly with another person through AID — both parties' AI assistants can naturally participate too.
Self-hosted home networks or team intranets: configuration, traversal, certificates, maintenance — each is an ongoing burden. AID-equipped devices connect directly with encryption, no public IP, no VPN server needed; hand connection management to your AI, and your private network becomes maintenance-free.
Share files, data, and services between multiple devices — flowing directly between them, no cloud storage platform in the middle.

All of these capabilities build on the same foundation: a self-sovereign identity, and the direct connections it makes possible.

Give your AI an address — get started Read the A2AL protocol docs