Local Mode

Your hardware.Your data.Free forever.

Run AGI entirely on your own machine with Ollama and LM Studio, or on-device on your phone. Local chats, files, and sessions never silently leave your device. No account required. Works offline.

Get AGI See AGI Desktop Explore the CLI

On-device
Works offline
No account required

Where Local runs

Three surfaces run fully local.

DesktopAGI DesktopOllama and LM Studio models in the native app. Chat, files, and tools without a cloud call.

TerminalAGI CLIThe agi binary works offline with local models. Sessions, code review, and sandboxed execution included.

MobileAGI MobileLocal Mode keeps conversations, memory, and files on the phone by default.

Verified commands

From local server to first prompt.

Point AGI at a running Ollama or LM Studio server, scan for models, and start working. The CLI infers the provider where it can, or you pass it explicitly.

agi · local mode

# With an Ollama server running
$ ollama pull <your-model>
$ agi models scan
$ agi --provider ollama --model <model>

# With an LM Studio server running
$ agi models scan
$ agi --provider lmstudio --model <model>

The boundary

Local is a trust boundary, not a toggle.

Private by boundary
In Local Mode, prompts, files, and responses stay on the device. Moving work to BYOK or Cloud is an explicit, labeled continuation. Never a silent handoff.
Free forever
Local inference runs on your hardware. No meter, no account, no cloud bill. Usage is bounded only by your device.
The full product shell
You get the same composer, projects, artifacts, model selector, and memory controls as every other mode, before any key or invite.

When you want more

Reach past your hardware, deliberately.

Bring your own provider keys on Desktop and CLI, or join the AGI Cloud waitlist for managed compute by invite. Either way, the move out of Local is an explicit choice with the provider label visible before anything is sent.

Set Up BYOK

Local Mode

Own the whole stack.

Follow the Desktop and CLI pages for current install routes, point AGI at Ollama or LM Studio, and keep your work on your machine.

Get AGI See AGI Desktop

Public launch · July 12, 2026

Your hardware.Your data.Free forever.

Run AGI entirely on your own machine with Ollama and LM Studio, or on-device on your phone. Local chats, files, and sessions never silently leave your device. No account required. Works offline.

On-device

Works offline

No account required

From local server to first prompt.

Point AGI at a running Ollama or LM Studio server, scan for models, and start working. The CLI infers the provider where it can, or you pass it explicitly.

agi · local mode

# With an Ollama server running
$ ollama pull <your-model>
$ agi models scan
$ agi --provider ollama --model <model>

# With an LM Studio server running
$ agi models scan
$ agi --provider lmstudio --model <model>

Local is a trust boundary, not a toggle.

Private by boundary

In Local Mode, prompts, files, and responses stay on the device. Moving work to BYOK or Cloud is an explicit, labeled continuation. Never a silent handoff.

Free forever

Local inference runs on your hardware. No meter, no account, no cloud bill. Usage is bounded only by your device.

The full product shell

You get the same composer, projects, artifacts, model selector, and memory controls as every other mode, before any key or invite.