Kimari Local AI

community

https://github.com/smouj/kimari-local-ai

AI & ML interests

Local AI for older NVIDIA GPUs. Open-source framework for running GGUF LLMs locally with llama.cpp, CUDA, GPU-aware profiles, CLI tooling, KimariFit scoring, and local OpenAI-compatible integrations.

Recent Activity

Smouj013 updated a Space 18 days ago

kimari-ai/README

Smouj013 updated a Space 18 days ago

kimari-ai/kimari-fit-lab

Smouj013 published a Space 21 days ago

kimari-ai/kimari-fit-lab

View all activity

Organization Card

Community About org cards

Kimari AI

Current framework version: v0.1.83-alpha Status: alpha framework, public model gate BLOCKED

Kimari Local AI — Local AI for older NVIDIA GPUs.

Kimari Local AI is an open-source, local-first framework for running GGUF language models on consumer NVIDIA GPUs. It combines llama.cpp CUDA inference, a practical CLI, GPU profiles, local configuration helpers, and an OpenAI-compatible endpoint for tools such as Open WebUI, OpenClaw, and Continue.dev.

Kimari is the framework. Kimari-4B is not released yet.

What works today

Area	Status
GTX 1060 local runtime validation	✅ Validated with a TinyLlama test model
llama.cpp CUDA runtime	✅ Validated locally
Local OpenAI-compatible endpoint	✅ `/v1/models` and `/v1/chat/completions` validated
Open WebUI integration docs	✅ Available
OpenClaw integration docs	✅ Available
Continue.dev integration docs	✅ Available
HF Jobs GPU smoke	✅ Completed
Private micro SFT pipeline	✅ Completed privately
Private adapter persistence	✅ Completed in a private repo
KimariEval Private v1	✅ Created, 104 private cases across 7 categories
Subset10 private completion/integrity eval	✅ Completed, not scored, manual review required

What is not released

The following are not public:

Kimari-4B public weights
public Kimari-4B adapters
official Kimari-4B GGUF files
public benchmark claims
production-ready guarantees

Any reference to Kimari-4B is a roadmap or private-pipeline reference unless explicitly stated otherwise.

Quick start

# Check local environment
kimari doctor --deep

# Download the small test model
kimari pull test

# Start the local endpoint with the test profile
kimari start --profile test

Then connect local tools to: