Extension
Extension runs local models on Apple Silicon Macs and makes them available to applications and coding agents through an OpenAI-compatible API.
Install
Requires macOS on Apple Silicon and Python 3.12.
python3.12 -m pip install hawking-runtime
hawking init
hawking init checks the available compute paths, installs a verified, pinned llama.cpp build for your user, and sets up supported coding agents it finds when run inside your project. It doesn't change your existing agent profiles.
The PyPI package is hawking-runtime, but the command is hawking. Do not install hawking from PyPI; that is a different project.
Run a model
hawking add Qwen/Qwen3-0.6B-GGUF:Q8_0
hawking serve MODEL
Use the library ID printed by hawking add in place of MODEL, or supply a local .gguf path.
hawking serve starts on 127.0.0.1:8277, or the next available port, and reports its choice of backend, context and available slots. These are based on the memory the machine has free at the time.
Your OpenAI-compatible client can use:
- Base URL:
http://127.0.0.1:8277/v1 - Model:
hawking - APIs: Chat Completions, streaming, tool calls and
/v1/responses
Compatible OpenAI clients, including Open WebUI when it can reach the local server, can use this endpoint. Check hawking status for the actual port if the default is occupied.
hawking serve MODEL --dry-run # inspect the plan without starting
hawking status
hawking bench
hawking stop
Other engines and devices
llama.cpp is the default. An experimental MLX adapter is also available for supported Apple Silicon configurations. You can register an existing engine on another host using an authenticated HTTPS connection:
hawking add https://HOST:8443
Hawking can route requests to registered compatible engines according to their measured capacity. Remote connections require the appropriate trust and authentication setup; see the coding-agent installation guide for details.
This Python package does not contain the NVIDIA driver or CUDA backend. Hawking eGPU is a separate component of the same Hawking Runtime project, distributed as developer source. Its current public compute capabilities and limitations are documented in the eGPU README.
Set it up with an agent
We encourage using a coding agent to set up Extension, especially if you're connecting your existing tools or working with a less common model or hardware configuration.
You can start with this prompt:
Read the Hawking Runtime Extension README and
AGENT_INSTALL.md. Help me install Hawking on my Apple Silicon Mac, inspect its available resources, and run a local model through the OpenAI-compatible API. Start withhawking init --dry-run, follow the documented installation and verification steps, then connect my supported coding agents without overwriting their existing configurations. Explain any compatibility or memory limits. Ask before installing software or changing system settings, and don't bypass security or verification checks.
For the exact agent procedure, including remote engines, verification, troubleshooting and rollback, read AGENT_INSTALL.md. To add devices, engines or preferences without modifying the package, use the AGENT_EXTEND.md guide whose location hawking init reports.
Compatibility and limits
The supported local configuration is an Apple Silicon Mac with Python 3.12. Linux may qualify in the future but we intend to maintain Extension and further Runtime within Apple's ecosystem.
The local server binds to loopback, not the public network. Hawking will refuse a model when its memory plan doesn't fit; free memory, choose a smaller model, or inspect the plan before retrying.
Licence
Copyright 2026 Hawking Systems Inc. See LICENSE, NOTICE and THIRD_PARTY_NOTICES.md for applicable terms and third-party attributions.
Metadata
Release files for hawking-runtime 0.1.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| hawking_runtime-0.1.0-py3-none-any.whl | Python 3 | none | any | Details for hawking_runtime-0.1.0-py3-none-any.whl |
Release files / hawking_runtime-0.1.0-py3-none-any.whl
| Download URL | hawking_runtime-0.1.0-py3-none-any.whl |
|---|---|
| Size | 518.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
941713d6659004ea2720f7b7bc04a34ada4408bbaf89c4d888e4e897618f3f19
|
|
BLAKE2b-256 checksum How to use checksums |
4c68698227bf92901338f39150681ada003f148446d2b0a76aadd3ed2de44717
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.6
|
Security for hawking-runtime 0.1.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.