How it works
Create the network
The owner establishes the private network and controls its membership.
Approve devices
Trusted machines join as Neurons that can offer models or agent capabilities.
Select private execution
A user chooses the Private Network for a task that should run beyond one machine but remain within the approved group.
Route encrypted traffic
Internet traffic can use Local AGI's relay to traverse NAT while remaining end-to-end encrypted between endpoints.
Review the result
The requester receives the artifact and verifies important output before acting.
What the relay does—and does not do
Local AGI's Private Network is not described as pure peer-to-peer. Devices on separate home or office networks often cannot accept direct inbound connections. A relay helps them connect across the internet.
The relay routes end-to-end encrypted traffic. It is not the machine that performs the AI task and should not receive readable task contents. Approved endpoint devices remain inside the trust boundary and must be protected accordingly.
Encryption protects the connection. It does not replace device security, careful membership approval, backups, permission controls, or output review.
Where a private network helps
- A team wants to use a workstation with a stronger GPU without uploading source files to an unknown public provider.
- A household has different devices with different models and wants to share capability.
- A lab or agency needs an explicit approved-device boundary for experimental workflows.
- Several machines must combine their memory to run one larger model as a hive.
- A user wants local-first control but occasionally needs intelligence unavailable on one machine.
Choose the smallest sufficient boundary
| Level | Use it when | Do not assume |
|---|---|---|
| Local AI | The task should remain on one device. | That every optional tool is automatically offline. |
| Private Network | Approved devices need to share intelligence. | That encryption makes an insecure endpoint safe. |
| Public Network | A general task can use broader available providers. | That confidential data is appropriate for public execution. |
Understand before connecting
Read the detailed guide to private AI network architecture, including trust boundaries and relay behavior. Learn how multiple devices run one model in the AI Hive Mind guide. If you are deciding between architectures, compare local and cloud AI agents.
Start on one device. Expand deliberately.
Use Local AI first, then create a Private Network when a task genuinely benefits from another approved machine.