Native Apple Silicon runtime
The homepage positions BaseRT as an LLM runtime built for Apple Silicon, with research material describing it as a native Metal inference runtime.
BaseRT
BaseRT is an LLM runtime for Apple Silicon Macs that runs local models on your own device for on-device inference and a local coding-agent workflow.
Overview
BaseRT is Base Compute’s LLM runtime for Apple Silicon. The product page describes it as the fastest runtime on Apple Silicon and shows a one-command install flow for running local models on your own device.
The public materials frame BaseRT around on-device inference: serving models locally, using them with a coding-agent plugin, and keeping model execution on your machine rather than sending requests to an external API. The research pages describe BaseRT as a native Metal runtime designed to improve throughput on Apple hardware.
The site’s research papers claim BaseRT supports a wide range of model families and multiple quantisation formats, but the product page itself is intentionally narrower and mainly highlights the Apple Silicon runtime and local-agent workflow. That makes BaseRT most clearly suited to developers and teams who want local LLM execution on Mac hardware.
Base Compute’s enterprise material places BaseRT inside a broader on-device AI stack that includes apps, control, routing, policy, and monitoring for organisations running AI on their own devices. The public page does not provide pricing details or a complete product guide, so the best-supported view is that BaseRT is the inference engine at the core of a local-first LLM setup.
Core capabilities
Performance-focused inference engine
A research report says BaseRT uses chip-specific kernel fusion, unified memory-aware optimisation, and custom dispatch logic to improve throughput on Apple hardware.
Local coding-agent workflow
The homepage provides a local coding-agent flow: serve a model, install the plugin, and keep the agent running on your machine without API keys or data leaving the device.
Multi-family model support
The source lists supported model families including Qwen, Llama, Gemma, Mistral, Phi-3, and Nomic BERT, indicating use across several open model lines.
Fits into a managed on-device AI stack
The enterprise page says the broader Base Compute stack includes model distribution, local/cloud routing, role-based access, policy enforcement, and monitoring, though not all of these are documented specifically for BaseRT on the public product page.
Common use cases
Local model inference on Apple Silicon
Run an LLM on a Mac so prompts and outputs stay on the local machine, which is the core workflow the homepage highlights.
Private coding-agent setup
Serve a local model and point a coding agent at it, using the plugin flow shown on the product page for on-device coding assistance.
On-device AI for organisations
Use BaseRT in a team or organisational context where the enterprise page emphasizes local-by-default deployment, no external API dependency, and data staying on owned devices.
Runtime evaluation and benchmarking
Evaluate BaseRT as an inference engine for open models when comparing it against other Apple Silicon runtimes, using the research pages and benchmark claims as the main evidence.
Pros and Cons
Pros
- Built for Apple Silicon, with research material describing native Metal execution.
- Designed for local inference, so model execution can stay on the user’s device.
- Supports a local coding-agent workflow without API keys in the example shown on the homepage.
- Covers multiple open model families across the site’s published examples and research.
- Backed by technical reports that explain the runtime’s optimisation approach and measured throughput claims.
Cons
- The public product page gives only a minimal setup example, so detailed installation, configuration, and supported-environment guidance are limited in the collected sources.
- Pricing is not available from the collected pages; the pricing URL returned a 404 in the provided evidence.
- The homepage lists model families, but it does not document every supported model, quantisation option, or compatibility edge case.
FAQ
What is BaseRT used for?
BaseRT is presented as an LLM runtime for Apple Silicon. The site positions it for running local models on your own device, including with a local coding agent workflow.
How do you install BaseRT?
The homepage shows an install command using curl and a shell script, but the site does not provide fuller setup documentation in the collected text beyond that one-command install example.
Which model families are supported?
The source says BaseRT supports model families including Qwen3, Qwen3.5, Qwen3.6, Llama 3.1, Llama 3.2, Gemma 3, Gemma 4, Mistral, Phi-3, and Nomic BERT.
Is there pricing information available?
The enterprise page describes BaseRT as local by default, with cloud only where you choose. The public pages do not provide pricing details, and the pricing page returned a 404 in the collected sources.
What integrations are documented?
The collected sources show BaseRT, technical reports, GitHub, Discord, and a coding-agent plugin example. They do not document broader third-party integrations or API details in depth.
Quick Facts
- Category
- Developer Tool / AI Infrastructure
- Platform
- Apple Silicon Macs
- Primary use
- Local LLM runtime for on-device inference
- Example workflow
- Serve a model locally, connect a coding agent, and keep execution on-device
- Source domain
- basecompute.co
- Pricing
- Not published in the collected sources
BaseRT Alternatives
ByteAsk
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repos and verifies changes with compilers, debuggers, sanitizers, and tests.
CreateOS Sandbox
CreateOS Sandbox is an isolated compute environment for running code and agent workloads in Firecracker micro-VMs with private networking and SDK, CLI, or MCP control.
hob
hob is an independent workspace for coding agents, with local control over sessions, terminals, history, routing, and follow-up work.
Manta AI
Manta AI is an autonomous web app testing tool that maps app behavior, catches regressions, and generates tests from a URL, no scripts or selectors needed.
SonOf
SonOf connects to your repo and PM tool, audits your codebase, and turns approved work into shipped tickets with senior engineering review.
Ghost
Ghost is a terminal-based AI assistant for chatting, code generation, and CLI tasks. Includes free models, supports Linux, macOS, Windows, and is open source.