BRIDGE1Have an invite code?
Your download should start now.
· Version
Download againSHA-256 checksumThe download link works for 15 minutes.
BRIDGE1 · MACOS EARLY ACCESS
Get local inference for vibe coding on macOS.
Install Bridge1, choose a coding model and copy your client setup for local inference on macOS.
Install Bridge1, choose a coding model and copy your client setup. The app prepares MLX or llama.cpp and checks memory before loading, so the first task is your code rather than engine flags. Version 0.7.8 is a universal disk image, signed with a Developer ID certificate, notarized by Apple and stapled. Apple Silicon is recommended. Intel is untested.
01 / DOWNLOAD
Get Bridge1
Early access is invite-only. Request an invite to replace recurring model-server setup with a managed local coding session.
No code yet?
02 / FIRST RUN
Install the inference engine. Let Bridge1 prepare the model.
- 1Install
Open the disk image and drag Bridge1 into Applications. Open it from there.
- 2Activate early access
Enter the invite code if the app asks for one. Activation sends a random device identifier and the code. No email address is needed.
- 3Add a model
Choose a model from the catalog. Bridge1 downloads its pinned files and prepares its engine.
- 4Connect your agent
On macOS, starting Bridge1 from Applications can set up the
bridge1command so you can start a coding tool withbridge1 launch. Copy the setup for your client from the Gateway panel. The Compatibility page has the steps.
03 / THIS RELEASE
Version 0.7.8
Bridge1 now tells you when another app is blocking it from starting, instead of closing silently.
Limits are listed in the app's Help, and per client on the Compatibility page.
04 / BEFORE YOU INSTALL
Check your Mac
24 GB of unified memory is the minimum. 32 GB or more is recommended.
Requirements →Choose your agent
OpenCode, Codex, Claude Code and Claude Desktop are tested end to end. Pi.dev and Hermes Agent have early support.
Compatibility →Know the limits
Some models are for light tasks only, and some clients have known limits. Each is listed on its page.
Models →BEFORE YOU USE BRIDGE1
Read the terms and the privacy policy.
The live Terms of Service and Privacy Policy on bridge1.ai are the operative text. Copies of them are on this site as drafts.
Read about Bridge1 and its publisher, Dojoit, Inc.
Terms of Service Privacy Policy →Download questions
Do I need an account to install Bridge1’s local inference engine?
No account is required for the local app flow; invite-only early access may require activation with your code. The installer and model downloads are separate steps. After installation, choose a catalog artifact so Bridge1 can prepare its engine instead of asking you to assemble a runtime.
Does installing Bridge1 make every coding client fully offline?
No. Local model inference keeps prompts and outputs on your computer, while activation, model downloads, updates and optional submissions need network access. Your coding client may also contact its own services. Check the client’s behavior instead of treating a local endpoint as an all-app offline guarantee.
Do I have to download every local coding model?
No. Choose the catalog model you need; Bridge1 downloads its pinned artifact and prepares the engine. Check quantization, disk size and memory fit first. The catalog includes MLX and GGUF profiles, so installing the desktop app does not mean loading every model into memory.
Do I need to install MLX or llama.cpp by hand?
No. Bridge1 prepares the engine for the chosen catalog model and verifies pinned downloads. Install the macOS app, activate if asked, add a model and copy your client setup. This replaces manual engine assembly; it does not remove model download time or memory requirements.
What should I check before redeeming a Bridge1 invite?
Check macOS availability, unified memory and the coding client you want to use. Early access is invite-only, with no published deadline or guaranteed acceptance. Choosing a model that fits and reviewing client limits avoids discovering those constraints in the middle of a coding session.