What you need
| Requirement | Details |
|---|---|
| Noema on both devices | Use Noema 4.1 or later on your Mac and iPhone. Updating only the Mac does not add Boost to an older iPhone installation. |
| Experience mode | Choose Simple or Advanced mode on both devices. Boost is unavailable in Beginner mode. |
| Compatible model | Load a supported Qwen 3.5 model in GGUF format on the Mac. Noema checks the model and its current runtime settings before preparing Boost. |
| Device connection | Use a USB data cable, or enable Wi-Fi on both devices. Direct Wi-Fi can connect the devices without a shared network. |
| Available resources | Keep enough free memory and storage on the iPhone for preparation. Finish its current model task before lending it to the Mac. |
The model still needs to load on the Mac. The extra context available from an iPhone depends on the selected model and available memory; free storage alone does not increase the context limit.
Use Installation for platform requirements and Downloading models to install a compatible model. A GGUF file alone does not guarantee Boost compatibility; follow the model choices and compatibility messages in the Boost panel.
Connect your Mac and iPhone
- Open Noema on both devices and load a supported Qwen 3.5 GGUF model on the Mac.
- Connect the iPhone directly to the Mac with a USB data cable, or enable Wi-Fi on both devices.
- On the Mac, open Constellation Boost from the model bar or Constellation settings.
- On the iPhone, open Stored → Constellation → Constellation Boost, then choose Use this iPhone.
- If Noema asks to unload the iPhone’s current model, review and confirm. Your saved chats remain available.
- On the Mac, choose Find connected iPhone. Select the iPhone and choose Connect when it appears.
- Compare the pairing code on both devices. Choose Codes match on each device only when the codes agree.
- Wait while Noema checks the connection and prepares the selected mode. For Faster prompts, the first preparation copies part of the model to the iPhone and keeps a cache for later connections.
- Keep Noema open on the iPhone and leave the devices connected while using Boost.
Choose a mode
| Mode | What the iPhone does | When to use it |
|---|---|---|
| Faster prompts | Processes part of a long prompt. The Mac continues generating the reply. | When prompt processing is the delay and Noema’s measured connection is suitable. |
| More context | Holds attention memory while the model stays on the Mac. | When Noema offers a larger context limit for your model and the iPhone’s available memory. |
Faster prompts
If the panel offers Prepare model for Boost, review the confirmation before continuing. Preparing this mode can reload the model with a temporary context limit. The dialog explains the change, and your saved model settings stay unchanged.
Prompt processing and writing a reply are different parts of inference. Sharing prompt work does not guarantee faster reply generation; replies can generate more slowly while Boost is on. Noema may report Your Mac is faster on its own when assistance is not worthwhile.
More context
Select an offered Context limit (tokens), then choose Use iPhone context. Noema calculates the available choices for the current model and iPhone. A smaller offered limit leaves the iPhone more memory.
Both devices must remain connected because the live conversation context now depends on the iPhone. Read the recovery steps below before disconnecting or turning off Boost.
Check whether Boost is helping
| Indicator | Meaning |
|---|---|
| Checking connection | Noema is measuring the route before deciding whether it can share work. |
| Boost ready | The devices are connected and preparation is complete. |
| Boost assisting | The model is sharing work with the iPhone. |
| Tokens processed together | Completed prompt work on the other device. This counter appears in Faster prompts after work completes. |
| iPhone context ready / Context held on iPhone | The More context profile is prepared or is holding live attention memory on the iPhone. |
The Link section shows the selected route, upload and download measurements, and latency. A ready connection confirms preparation; use the activity and completed-work indicators to see what happens during a request.
Choose a connection
- Direct cable: use a USB data cable connected directly to the Mac. Cable and port speed affect the measured link; a charging-only cable cannot carry the connection.
- Local network: enable Wi-Fi on both devices and allow the local connection. Network isolation or device permissions can prevent discovery.
- Direct Wi-Fi: a peer-to-peer route that does not need the devices to join the same network.
Noema measures available links and uses the fastest suitable route. If another route is available, the panel can offer Try cable or Try Wi-Fi. Keep the devices nearby and follow the measured link guidance rather than assuming a cable is always faster.
Keep a session running
Sessions last one hour. The iPhone shows Extend session near the end if you want to keep lending it to the Mac. Choose Turn off Boost when you are finished.
Locking the iPhone, switching Noema into the background, heat, low battery, or memory pressure can pause assistance. Keep Noema in the foreground and follow any cooling, charging, or memory guidance shown in the panel.
Recover a lost context connection
If a More context connection ends, the current reply stops and the chat remains saved. The panel offers recovery rather than continuing with missing context.
- Reopen Noema on the iPhone and reconnect the devices.
- Follow Reconnect to rebuild context to rebuild the conversation context from the saved chat.
- If the iPhone is unavailable, choose Return to Mac context to restore the Mac’s original context limit.
A conversation that needed the expanded limit may need to be shortened or summarized before it fits the Mac’s original limit. The phone’s live attention memory and your saved transcript are different resources.
Troubleshooting
| Message or symptom | What to do |
|---|---|
| No iPhone found yet | Check Noema’s version and experience mode on both devices. Open Boost on the iPhone, choose Use this iPhone, and check the USB data cable or Wi-Fi route. |
| Local Network access denied | Use Open Local Network Settings and allow Noema on the affected device. Return to the app and choose Retry. If access is already allowed, follow the panel’s instructions to toggle the permission and reopen that copy. |
| Load a compatible GGUF model | Select a supported Qwen 3.5 GGUF model on the Mac. Remote endpoints and models running in other formats do not satisfy Boost’s local GGUF requirement. |
| This model setup cannot use Boost | Review the model and current settings. Use Prepare model for Boost if offered, and read its reload and context-limit confirmation. |
| A faster connection is needed | Try the alternate cable or Wi-Fi route. Follow the link meter’s cable and port guidance. |
| Your Mac is faster on its own | Use the model on the Mac without assistance for this setup. Boost does not promise a benefit for every model, prompt, or connection. |
| Not enough available memory / storage | Free resources on the affected device or choose a smaller compatible model. For More context, use a smaller offered context limit. |
| Open Noema on your iPhone | Unlock the iPhone and keep Noema in the foreground. |
| Paused while your device cools down / Connect your iPhone to power | Let the device cool or connect it to power before retrying. |
| Workspace policy does not allow Boost | Check the managed workspace’s device and model permissions with its administrator. |
Use Forget device to remove a stored pairing, then repeat the code confirmation if you need to pair again. Clear Boost cache removes prepared Boost copies and can require another transfer during the next preparation.
If a problem persists, contact Support with the device versions, model name, selected mode, route, and exact message displayed by Boost.
Privacy and network activity
Boost pairs the devices with a code and encrypts the connection. Its pairing and shared inference stay between the paired devices and do not require internet or a cloud relay.
Model downloads, app updates, chat sync, Web Search, and configured tools retain their own network behavior and permissions. Enabling Boost does not make those other features offline. Read Privacy & Network Activity for those routes.