Inferencer lets you run, host and deeply control the latest SOTA AI models (OSS, DeepSeek, Qwen, Kimi, GLM, MiniMax and more) from your own computer.
No data is sent to the cloud for processing - maintaining your complete privacy.
Advanced inferencing controls give you complete control on their accuracy and outputs.
Models
Start in the models section where you can select the location of existing models or download new ones directly from Hugging Face.
Use the distributed compute feature to load a model across two Macs, or use the model streaming feature to inference larger models partially from storage.
Chats
Select the model to interact with on the top menu bar and write a prompt to begin. At any point you can switch between models and continue the chat to see what else they can uncover. You can also selectively delete past messages to keep the model focused and less scatterbrain.
Chat Controls
Control the inferencing parameters including batching to inference multiple chats at the same time, intensity of processing, context quantization to further reduce the memory usage and model streaming to load models larger than available memory.
Token Entropy and Inspection
Select the inspectors to peek into the inner-workings of each word outputted and see the model's confidence levels and alternative choices.
Response Control
Utilise the framing feature which allows you to control the output the model generates. For example, skipping the preamble or directing the model to output in structured html.
Tools
The tools editor allows you to enable built in tools such as get_webpage_content or add in your own, so that models can use them when needed. For example, if you'd like a webpage or search result inferenced, simply enable the tool in the Tools section, and allow tool calls in the chat settings panel.
Server
If enabled, the server feature allows you to serve and connect to your own or trusted devices. No data is sent elsewhere. Also includes compatible APIs for application development.
Prompt Caching
Significantly improves prompt processing by prefix-matching previously processed prompts. Uses a user-configurable cache pool with optional external storage and automatic LRU (least recently used) eviction (enabled in Settings).
Distributed Inference
With distributed compute you can link together two Macs, sharing the memory to inference larger models. To use make sure it's enabled in both the app and server settings. Once a connection to your server is made, if both the computers have the same model, a distributed compute icon will appear. Simply tap on it to load the model for distributed compute.
Coding Tools
Built-in support for Xcode Intelligence and Visual Studio Code. Use the server feature with Compatibility APIs enabled and SSL disabled to allow Xcode or Visual Studio Code to use Inferencer as a service provider.
Shortcuts
Use the Shortcuts app to automate inferencing workflows (e.g., copy text from clipboard > inference > speak result).
Settings
Includes parental controls, an automatic deletion policy and more.
Privacy
For maximum privacy, all AI processing happens offline and on your device, by default.
Subscriptions
Basic (Free): Most features unlocked for free including unlimited chats.
Professional: Upgrade for more advanced token inspection, prompt-framing and model streaming.
Terms & Support
Terms of Use: inferencer.com/terms
Privacy Policy: inferencer.com/privacy
Disclaimer
Inferenced models may not always be accurate or contextually appropriate. You are responsible for verifying the information before making important decisions.
App would not download models but the developer responded to email with a quick fix, and it should be fixed in next version. This is an interesting program - I haven’t figured out all the bells and whistles.
Developer Response
Hi, an update was submitted to Apple earlier in the week to address an issue with downloading on macOS 26. As a workaround, you can download the models directly from HuggingFace and place them directly in your chosen Downloads folder. Feel free to report any other issues or requests you have on the public roadmap.
Like nothing before
txg_sync
This app does something nothing else does: allow you to see alternative tokens in nearly real-time. And to adjust the output to your framing without complicated setup or per-model sidebar configuration. You can do it per conversation. For AI researchers, the curious, and those looking to tweak model outputs? This is a fabulous tool. Highly recommended for the curious and local LLM enthusiasts.
Developer Response
Thank you so much for the support and feedback, if there's any other features you'd like implemented feel free to add it to the public roadmap.
Probably good, but…
Jackdeath3
As someone who has multiple Macs, I was looking forward to using this for distributed inference, but I think that charging a monthly subscription is a really bad business model. I would love to support the development of this program, but monthly fees are the exact kind of thing that I wanted to avoid by investing in local AI hardware. I hope you consider changing this to a one time payment.
Developer Response
Hi Jack, thanks so much for sharing your perspective.
Most Quality Inferencing for OS X
packetssx
The maker of this app, xCreate is well beyond knowledgeable when it comes to inferencing LLM’s. Though some functions of using it as an agentic coding api endpoint needs work, the creator of this app is updating and fixing things on a literal daily basis. Keep in mind that he is allowing all of this to be used for free.
Developer Response
Thanks so much for the support. If there's any specific issues you have with the api endpoints, please post them on the issues page or send over an email and we'll get it fixed for you.
+ Support for M5 Neural Accelerators (~4x prompt processing)
+ HTML block support
+ ModelScope downloader
+ Patches for DeepSeekv32, Step35 and Distributed Compute
+ Improved support including crash fixes for macOS 26.4
+ More bug fixes and performance improvements
Also in case you missed the last update:
+ Support for Gemma4, Qwen 3.5, Nemotron Super, Mistral Small 4, Bonsai, Sarvam, Ring-2.5-1T
+ Improvements to prompt caching (reduced cache misses, 99x faster when cached)
+ Support for TurboQuant (Unbatched) in Model Context Precision
+ Auto-launch on login, hide Dock and Menu Bar options in Settings
+ Prompt caching and Tool call support for Distributed Compute
+ Download models filter (including filter by system memory)
+ Pro Yearly plan (2 months free)
Version v1.10.12
The developer, Ashraf Samy, indicated that the app’s privacy practices may include handling of data as described below. For more information, see the developer’s privacy policy .
Data Not Collected
The developer does not collect any data from this app.
Privacy practices may vary, for example, based on the features you use or your age. Learn More
Accessibility
The developer has not yet indicated which accessibility features this app supports. Learn More
Information
Seller
Ashraf Samy
Size
162.7 MB
Category
Productivity
Compatibility
Requires iOS 18.0 or later.
iPhone Requires iOS 18.0 or later.
iPad Requires iPadOS 18.0 or later.
Mac Requires macOS 15.0 or later and a Mac with Apple M1 chip or later.
Apple Vision Requires visionOS 2.0 or later.
Languages
English
Age Rating
9+
9+
This app has an age rating of 9+ with content restrictions. Some content may be rated higher, but access is managed by the developer through in-app controls.
