--- license: mit base_model: deepreinforce-ai/Ornith-1.0-9B pipeline_tag: text-generation library_name: gguf language: - es - en tags: - gguf - llama.cpp - ollama - personnnos - function-calling - local-ai - agent - tool-use - conversational --- # Personnn Buddy 9B GGUF Buddy is PersonnnOS's local language model. It is tuned for conversational assistance, tool use and desktop workflows in Spanish and English. This repository contains the `Q4_K_M` GGUF build for local inference with PersonnnOS, llama.cpp or Ollama. > **🐼 Give Buddy a body.** This is just the brain. PersonnnOS is the sovereign > desktop app where Buddy lives: a native browser, workspace, tools and agent — > all on your machine, no cloud account required. It installs and runs this > model for you with its bundled llama.cpp runtime. > > **→ Download PersonnnOS (free) at [personnn.com](https://personnn.com)** ## PersonnnOS in action Buddy runs inside PersonnnOS — a native browser, workspace and agent on your machine. | | | |---|---| | ![Buddy in PersonnnOS](media/buddy-studio.png) | ![Private inference on your VPS](media/remote-inference.png) | | **Buddy, local and adaptive** — presence, appearance and behavior under your control. | **Private AI on your VPS** — run Buddy on your own infrastructure; data never touches third parties. | | ![Brain context graph](media/brain.png) | ![Personal Life preview](media/personal-life.png) | | **Brain** — reasons over your project's files, relations and memory. | **Personal Life (preview)** — your agent lives in a shared city: journals, building, trade. | ## Model details | Property | Value | |---|---| | Base model | [deepreinforce-ai/Ornith-1.0-9B](https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B) | | Architecture | Qwen3.5, 9B parameters | | Fine-tuning | LoRA/SFT for PersonnnOS workflows | | Training examples | 397 curated examples | | Quantization | Q4_K_M | | File size | 5.24 GiB | | Recommended context | 12,288 tokens | | License | MIT | ## Download Download `personnn-buddy-9b-v2-Q4_K_M.gguf` from this repository. PersonnnOS can install and run it with its bundled llama.cpp runtime, without Ollama or a cloud account. Verify the file after downloading: ```bash shasum -a 256 personnn-buddy-9b-v2-Q4_K_M.gguf ``` Expected SHA-256: ```text 888bb1b58795277abc2371065f88f6f03e4777d881385720416e328ba14809fa ``` ## llama.cpp ```bash llama-server \ -m personnn-buddy-9b-v2-Q4_K_M.gguf \ --ctx-size 12288 \ --flash-attn on \ --host 127.0.0.1 \ --port 8080 ``` ## Ollama Keep `Modelfile` and the GGUF in the same directory, then run: ```bash ollama create personnn-buddy:9b-v2 -f Modelfile ollama run personnn-buddy:9b-v2 ``` The included Modelfile defines the intended Buddy identity, context length and sampling defaults. ## Evaluation Buddy v2 scored **11/12 (92%)** in PersonnnOS's internal workflow evaluation, with an average first-token latency of **6.9 seconds** on the test machine. The suite covers chat, structured tool selection and desktop tasks. These results are internal product measurements, not a general-purpose benchmark. ## Intended use - Local conversational assistance. - PersonnnOS tools and desktop workflows. - Spanish-first personal productivity. - Private on-device inference when run through a local runtime. The model does not itself enforce permissions. PersonnnOS places tool calls behind its permission broker and requires confirmation for consequential actions. ## Limitations - The model can hallucinate or select an incorrect tool. - Tool availability and schemas are supplied by the host application. - Outputs are not professional legal, medical or financial advice. - External, destructive or irreversible actions should require explicit user confirmation. - Local privacy depends on the runtime and application configuration. Using a cloud-hosted runtime sends prompts to that service. ## Provenance Personnn Buddy 9B is derived from [Ornith-1.0-9B](https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B), published by deepreinforce-ai under the MIT license. See `THIRD_PARTY_NOTICES.md`. ## About PersonnnOS PersonnnOS is a privacy-first personal agent and browser. Learn more at [personnn.com](https://personnn.com).