Local

Product Introduction

1. Definition: Local is a native macOS application (with Windows and Linux versions in development) that functions as a fully private, on-device AI inference platform. It is a technical suite that bundles a curated model catalog, a proprietary inference engine (BaseRT), and a user-friendly interface to run large language models (LLMs) and AI tasks directly on a user's personal computer hardware.
2. Core Value Proposition: Local exists to provide a zero-cost, zero-setup, and completely private alternative to cloud-based AI services. Its primary value is enabling users to leverage powerful AI for chat, coding assistance, and meeting transcription without data ever leaving their device, addressing critical concerns around privacy, data sovereignty, and ongoing API costs.

Main Features

1. Privacy-First Architecture: All AI processing occurs locally on the user's Mac. The app includes a dedicated "Privacy Mode" which ensures no prompts, generated content, or contextual memories are transmitted to external servers. User data, including conversation history and a manually editable "Memory" feature, is stored in a local database on the device.
2. Self-Optimizing Performance with BaseRT: The app ships with BaseRT, Basecompute's proprietary inference engine. Unlike generic builds, BaseRT is compiled specifically for the user's Apple silicon chip (M1, M2, M3, etc.) upon first launch. This hardware-aware tuning is claimed to deliver performance gains of up to 5.4x compared to standard engines used in other local AI applications.
3. Hardware-Adaptive Model Management: Local automatically audits the host Mac's unified memory capacity and processor. It then recommends and manages a selection of open-source models (like Llama, Mistral, Gemma) that optimally fit the available resources. This allows a MacBook Air with 8GB RAM to run efficient models, while a Mac Studio with 128GB RAM can leverage larger, more capable models.
4. Office Mode (Local Network Server): This feature allows one powerful machine (e.g., a Mac Studio, a desktop PC, or an on-premise server) to act as a private AI server on the local network. Other devices (laptops, other Macs) running the Local client can then connect to it, enabling resource sharing and access to larger models without needing powerful hardware on every endpoint.
5. Integrated Multi-Modal Workspace: The app provides a unified interface for distinct AI tasks: a Chat interface for document Q&A (PDFs, text), a Coding agent that can read, edit, and execute code within a local project folder, and a Meetings tool that performs on-device speaker-labeled transcription using models like Whisper.

Problems Solved

1. Pain Point: Escalating costs and lack of control associated with proprietary cloud AI APIs (OpenAI, Anthropic). Users pay per token, leading to unpredictable expenses, and have no guarantee of data privacy as all inputs are processed on external servers.
2. Pain Point: The complexity and technical barrier to entry for running open-source LLMs locally. Users often struggle with manual setup, environment configuration, model sourcing, and performance tuning for their specific hardware.
3. Target Audience: Privacy-conscious professionals (lawyers, healthcare workers, journalists), developers and engineers seeking offline coding assistance, businesses requiring data sovereignty (legal, financial, government sectors), and cost-sensitive individuals or teams wanting to experiment with AI without subscription fees.
4. Use Cases: Drafting sensitive documents with AI without exposing content; getting coding help within a proprietary codebase that cannot be uploaded to the cloud; transcribing confidential internal meetings on-device; providing a shared, cost-free AI resource for an entire small office via Office Mode.

Unique Advantages

1. Differentiation vs. Cloud APIs: Unlike OpenAI's ChatGPT or Anthropic's Claude, Local requires no account, has zero per-token cost, and ensures complete data privacy by keeping all information on-device. It trades access to the very latest "frontier" models for total control and ownership.
2. Differentiation vs. Other Local AI Apps: Compared to other local AI applications, Local's key innovation is the BaseRT inference engine and its hardware-specific compilation. While others may ship a one-size-fits-all backend, Local's automatic tuning for the user's specific Apple silicon chip is a claimed unique performance advantage. Its integrated, task-specific workspace (Chat, Coding, Meetings) also offers a more polished user experience than generic chat interfaces.
3. Key Innovation: The integration of a hardware-aware, self-optimizing stack. The combination of automatic hardware detection, model recommendation, and a chip-compiled inference engine (BaseRT) removes the technical friction typically associated with local AI, delivering a "it just works" experience tailored to the user's specific Mac.

Frequently Asked Questions (FAQ)

1. Does Local AI work completely offline? Yes, Local is designed for full offline operation. After the initial download of the application and your chosen AI models, all processing—including chat, coding, and meeting transcription—occurs entirely on your Mac without requiring an internet connection.
2. What are the system requirements for running Local on Mac? Local requires a Mac with Apple silicon (M-series chips). The key limiting factor is Unified Memory (RAM). The application will automatically recommend models suitable for your available memory, with 8-16GB supporting smaller models, 24GB enabling mid-size models, and 32GB+ providing the best experience with larger, more capable open-source LLMs.
3. Is my data safe and private with Local? Absolutely. When using Privacy Mode, no data—including your prompts, generated responses, uploaded files, or meeting audio—ever leaves your computer. Everything is processed locally and stored in a database on your Mac. The company states there is no telemetry or data collection.
4. Can I use cloud AI models like GPT-4 alongside local models in Local? Yes. Local offers an optional "Cloud" section where you can configure your own API keys for services like OpenAI, Anthropic, or OpenRouter. This allows you to access frontier models when needed, but these requests are explicitly routed to the provider and are disabled when Privacy Mode is active.
5. How does Office Mode work and what can I use as a server? Office Mode turns one machine on your local network into a private AI server. You install Local on a powerful machine (e.g., a Mac Studio, a Windows/Linux desktop with a high-end GPU, or even an on-premise server). Other devices on the same network can then connect their Local client to this server's IP address, allowing them to use its computational power and access larger models it hosts.