Skip to content
runa

Local GGUF · OpenAI · Anthropic

compute local

Local fire.
Cloud when you need it.

runa is a single CLI that runs AI models locally via GGUF / llama.cpp, or through OpenAI and Anthropic APIs — same thinking controls, tools, and serve surface.

View on GitHub → runa fit

One binary, two worlds

local

GGUF on your machine

Run ggml / llama.cpp models with cpu, gpu, or hybrid placement. Audio and vision inputs when the model supports them. Adaptive memory shrinks when idle.

cloud

OpenAI & Anthropic APIs

Same deep-thinking controls, structured output, and tool calling across local and cloud. OpenAI-compatible runa serve for editors and agents.

runa fit

Know if it fits before you download

Forecast memory and speed with a confidence interval, calibrated by real runs. runa fit --recommend ranks a curated catalog for this machine.

runa fit hf:unsloth/Qwen3-30B-A3B-GGUF:Q4_K_M --ctx 16384 --kv q8_0
runa fit --recommend --use code

Capabilities

01 / think

Deep-thinking control

off / on / token budget / effort — same knobs for local and cloud models.

02 / serve

OpenAI-compatible server

runa serve loads the default model at startup, reports progress on /health, and waits instead of failing mid-load.

03 / tools

Structured + MCP

JSON Schema or GBNF grammar, tool calling on serve routes, MCP tool loop in run / chat.

Open source

Run models where they belong.

Local first, cloud when the job needs it — one CLI, one config surface.

Star on GitHub →