This is a common misconception because most on-device inference frameworks in the market primarily target the CPU or GPU of the user's device. In order to avoid slowdowns for demanding non-inference workloads such as video conference meetings and games, on-device inference should avoid the CPU and GPU.

In sharp contrast, on-device inference with Argmax SDK targets the NPU (Neural Processing Units) and avoids resource contention with non-inference workloads. For example, running WhisperKit (NPU) while attending a Zoom meeting (CPU+GPU) should not slowdown the Zoom meeting.

TL; DR: If you observed slowdowns in the past with others, try again with Argmax SDK! You should not observe a slowdown. If you do, please contact us on Discord we can help troubleshoot

Basic Plan can be used without internet connection after the initial download of the model files.
‍
‍Pro Plan requires internet connection only once every 30 days in order to renew the device license.
‍
Enterprise Plan may request custom connectivity requirements (including fully offline).

In sharp contrast, all server-side and most on-device inference providers require internet connectivity for each and every inference request.

Per-device pricing covers a single device that your app is installed on and where Argmax SDK gets initialized at least once during that billing month.

Per-user pricing covers aN active single user of your app. To enable per-user pricing, Argmax SDK requires the developer to pass a unique (non-tracked) user identifier during SDK initialization for billing purposes. There is a maximum provisioning ratio of 3:1. If the deployment exceeds the 3:1 maximum provisioning ratio, the device overages will be billed using the per-device pricing.

For example, an application with 1000 unique reported users may have up to 3000 device licenses active and pay for 1000 per-user licenses. If the actual active device license count is 4000, then the developer will also be billed for 1000 per-device licenses.

Pro Plan is priced per-device or per-user per month with unlimited usage, making your costs predictable. In sharp contrast, server-side inference providers follow usage-based pricing, introducing unpredictability.

‍Enterprise Plan includes volume discount tiers with significantly reduced unit costs at scale.

For each end-user device with a valid license: There is no rate limit. There is no concurrency limit. There is no usage volume limit.

Basic Plan and Pro Plan support iOS, iPadOS, macOS.
Enterprise Plan additionally supports Android and Linux.
Windows support is on our roadmap.

The Argmax SDK is continually benchmarked on a large fleet of real devices. These results are published publicly on the WhisperKit Benchmarks Gradio space and updated regularly with new versions of the SDK.

As new devices are introduced to the market, we systematically expand our testing fleet to ensure that we remain current with the latest performance metrics across device categories. This enables developers to validate the SDK's performance against their specific user base.

Argmax is committed to maintaining open-source projects such as WhisperKit to increase the adoption of FMOD technology by everyone.

Argmax is also committed to serving its customers with cutting-edge performance improvements and advanced features built on top of open-source exclusively as part of Argmax SDK.

Argmax follows an open-core approach and parts of the Argmax SDK may be open-sourced over time.

Argmax believes that privacy is a fundamental human right, so every Argmax product and service is designed to minimize the collection and use of your data and use on-device processing whenever possible.

Pro Plan end-user devices may communicate with our API for software licensing and performance telemetry purposes. The data schema is available to active subscribers of the Pro Plan upon request.
‍
‍Enterprise Plan may request custom behavior such as air-gapped deployment with zero data collection.

Basic Plan is best for academics and hobbyists to tinker with, build on top of and/or even commercialize open-source Argmax FMOD technology. The MIT license is friendly to this cohort.

We are actively working on real-time text-to-speech and language model framework products. Please inquire at info@argmaxinc.com if you have a use case or a roadmap request!