# PrismML-Eng/Bonsai-demo

Bonsai Demo

Repository: https://github.com/PrismML-Eng/Bonsai-demo
Canonical: https://ross.abutalabs.com/products/bonsai-demo
Homepage: https://prismml.com/
Language: Shell
License: Apache-2.0
License Family: permissive
Topics: bonsai, llm, prism-ml, llamacpp, mlx, small-models
Last push: 2026-08-25T19:20:55+00:00

## Health v2 (maintenance only)
Score: 59/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 35, longevity 11
- inputs: {"age_days": 161, "days_push": 8, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2252, forks 229 (observed 2026-08-28T04:06:31.066405+00:00)

## What it is
A demo repository for running PrismML's Bonsai family of 1-bit and ternary-quantized language models locally via llama.cpp and MLX. It provides setup scripts and a local chat server (with vision, tool calling, MCP, and long-context support) for models ranging from 1.7B to 27B parameters.

## Use cases
- run a 27B LLM locally on a laptop or iPhone
- chat with a local vision-language model about screenshots and PDFs
- self-host an OpenAI-compatible LLM server with tool calling
- run an LLM offline with full data privacy
- test ultra-low-bit quantized models on Metal, CUDA, Vulkan, or ROCm
- build agentic workflows with a small local model and MCP servers

## When to choose
- you want a capable LLM running entirely on-device with a tiny memory footprint
- you need local inference on Apple Silicon, consumer GPUs, or CPU
- you want vision, reasoning, and tool calling without cloud APIs
- energy efficiency and token throughput on edge hardware matter

## When to avoid
- you need maximum output quality from frontier-scale models
- you require a managed cloud inference service
- you need fine-tuning or training rather than inference
- you depend on an ecosystem beyond the provided llama.cpp/MLX demo scripts

## Facets
- artifact type: application
- maturity: active
- function: llm-inference, chatbot, rag, mcp, machine-learning
- domain: large-language-models, artificial-intelligence, machine-learning, self-hosted, developer-tools
- platform: windows, cross-platform, cli, self-hosted
- tags: llamacpp, mlx, 1-bit-quantization, ternary-weights, local-llm, on-device-ai, vision-language-model, small-models, bonsai, prism-ml, macos, linux, gpu

## Member repositories
- PrismML-Eng/Bonsai-demo (main) score 59

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:31.066405+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:43:48.581453+00:00, confidence not recorded.
  - readme: https://github.com/PrismML-Eng/Bonsai-demo (fetched 2026-08-28T04:06:31.066405+00:00, sha 6efc2c93e197)
  - homepage: https://prismml.com/ (fetched 2026-08-29T10:23:53.788505+00:00, sha 4fcb67de1755)
  - site_page: https://prismml.com/about (fetched 2026-08-29T10:23:53.790846+00:00, sha 0a52b07a3d6d)
  - site_page: https://docs.prismml.com/ (fetched 2026-08-29T10:23:53.792587+00:00, sha be8a4ca6d7e8)
  - site_page: https://prismml.com/news/bonsai-27b (fetched 2026-08-29T10:23:53.794303+00:00, sha 055732c4c7fe)
- Data as of 2026-08-30T08:39:29.467469+00:00.