# llama.cpp v1.0.0 ships with stable API rules for builders

> llama.cpp v1.0.0 and ggml v1.0.0 are out. Major version bumps will now signal API and ABI breaks. Here is what that means if you build on local AI.

- URL: https://tools.launchllama.co/blog/llama-cpp-v1-0-0-stable-api-semver
- Published: 2026-11-13
- Publisher: Launch Llama (https://tools.launchllama.co)

TL;DR
- llama.cpp v1.0.0 (libllama) and ggml v1.0.0 (libggml) are released, announced on the Hugging Face blog on November 13, 2026.
- Breaking changes to the public C API will now trigger a MAJOR version bump. Features, backend updates and model support are MINOR. Bug fixes are PATCH.
- llama.cpp supports more than 150 architecture variants and is distributed via pre-release artifacts, Docker, Debian, Homebrew, conda-forge, MacPorts, Winget and XCFramework.

## What happened

The ggml-org team released v1.0.0 of llama.cpp (the libllama library, public headers in llama.cpp/include/) and of ggml (the libggml tensor library, headers in ggml/include/). The announcement came on the Hugging Face blog on November 13, 2026, about 3.5 years after the project started.

The headline change is a versioning promise. Both projects will reflect API and ABI incompatibilities with major-version bumps. The rules: a breaking change to the public C API in ./include is MAJOR, a backward-compatible feature, backend update, model support or API addition is MINOR, and a bug fix with no API change is PATCH. The team says it will be stricter about making those boundaries visible to downstream consumers as adoption grows.

The post also covers distribution and infrastructure. Listed channels are pre-release artifacts (Linux, macOS, Windows, Android, iOS), Docker, Debian, Homebrew, conda-forge, MacPorts, Winget and XCFramework (iOS, macOS). The team says it is investing in CI using HF Buckets and HF Jobs, with hardware support from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel for self-hosted runners. The post does not list specific code changes or a migration guide, so check the release notes for those.

## What this means for founders

- **Safer dependency.** If your product embeds libllama or libggml, a major-version bump now signals a breaking API or ABI change. You can pin to v1 and treat MINOR and PATCH updates as the backward-compatible ones.
- **Less surprise maintenance.** Clear semver rules should make upgrade planning easier. How strictly this holds in practice is not confirmed yet, since the post only states intent.
- **Wider reach for shipping.** The listed channels include iOS and Android artifacts, an XCFramework for iOS and macOS, and Winget, Homebrew and conda-forge packages. That covers the main platforms for cross-platform and private AI products.
- **Model coverage.** More than 150 architecture variants are supported, plus all mainstream speculative decoding methods. Check your target model is on the supported list before you commit.
- **Cost angle.** The source does not mention pricing. Running inference locally is the project's pitch for private products, but any cost savings versus a hosted API are not covered in the post.

## What the Launch Llama directory shows

Live numbers from Launch Llama for listings that describe themselves as "run llms locally", "local llm inference", "on-device ai" tools, as of 2026-11-13.

SignalLaunch Llama dataListings in this space4New in the last 30 days0With a free or freemium plan100%Most upvoted by foundersKrisp (396 upvotes)

Our directory has 4 matching listings in this space, none of which went live in the last 30 days, and 100% have a free or freemium plan, so it is a quiet, low-price corner right now.

## How to try it today

- Open the ggml-org projects on GitHub and Hugging Face and find the v1.0.0 releases for llama.cpp and ggml.
- Pick the distribution channel that fits your stack: pre-release artifacts, Docker, Debian, Homebrew, conda-forge, MacPorts, Winget or the XCFramework.
- If you already embed libllama, read the public headers in llama.cpp/include/ and the release notes to see whether anything affects your build.
- Set your dependency to track v1 so a future major bump does not break your build unnoticed.
- Check the backend list and the supported architectures for the hardware and model you plan to ship on.

## Related tools on Launch Llama

Tools from the Launch Llama directory that fit with this news:

ToolWhat it doesWhy it matters hereOsaurusFor Mac users: Osaurus runs AI agents 100% locally on your machine with persistent memory and cryptographic identity, fully offline and open source.Osaurus runs AI agents fully locally and offline on a Mac, the kind of local AI product category this release aims to support. The listing does not say it uses llama.cpp.OpenHumanFor privacy-conscious users: OpenHuman is a local-first, open source AI agent that remembers everything and improves over time.OpenHuman is a local-first, open source AI agent, an example of the privacy-focused local AI products the llama.cpp team says it expects to grow.MemoriaFor photographers: Memoria searches your camera roll by text, speech, objects and faces using on-device AI so you find memories instantly without scrolling.Memoria uses on-device AI to search a camera roll, showing a shipped consumer product built on local inference. The listing does not say it uses llama.cpp.

[Browse all tools](https://tools.launchllama.co/tools) or [list your own tool](https://tools.launchllama.co/submit).

## Llama's take

I would try this now if you already build on llama.cpp or plan a local or private AI feature. A stated versioning policy is what you want before depending on a C library in a product. The 1.0 label itself does not change the code, but the promise about breaking changes does matter.

I would wait if you only run models through a hosted API and have no reason to go local. Also wait on judging how well the policy holds until you see a few releases. The post gives intent, not a track record under the new rules.

The one thing to watch is the first major-version bump after this one, and how clearly it is communicated.

## FAQ

### What is llama.cpp v1.0.0?

It is the first major-version release of the libllama library, released alongside ggml v1.0.0 (libggml). The team announced it on November 13, 2026 to give downstream developers a flexible but stable API.

### What counts as a breaking change in llama.cpp now?

A breaking change to the public C API in the ./include directory triggers a MAJOR version bump. Backward-compatible features, backend updates, model support or API additions are MINOR, and bug fixes with no API change are PATCH.

### Where can I get llama.cpp?

The source lists pre-release artifacts for Linux, macOS, Windows, Android and iOS, plus Docker, Debian, Homebrew, conda-forge, MacPorts, Winget and an XCFramework for iOS and macOS. It also points to a discussion on adding ggml-org projects to packaging pipelines.

## Sources

- [llama.cpp v1.0.0 : A stronger foundation for the local AI ecosystem](https://huggingface.co/blog/ggml-org/llama-cpp-v1) (Hugging Face)

---

More for agents: https://tools.launchllama.co/llms.txt · https://tools.launchllama.co/catalog.md · https://tools.launchllama.co/mcp
