Open-source local LLM coding agent & SDK
OpenJet profiles your hardware, configures Qwen or another open model, and gives you a terminal coding agent plus a Python SDK — without sending your code or prompts to a hosted API.
pip install open-jetTwo ways to use it
OpenJet’s job is the part nobody wants to do: profiling the hardware, picking a model that fits it, and configuring the local runtime. What you point at that runtime is up to you.
Terminal agent
Install it, run setup, and open the agent. It reads and searches files, applies edits, resumes sessions, and asks before it runs anything that changes state.
pipx install open-jet
openjet setup
openjetPython SDK
Same harness, no terminal. Open an inference session and call it — the model runs on the device your code is already running on.
from openjet.sdk import create_inference_session
session = await create_inference_session(
system_prompt="You are a concise shopkeeper.",
)
reply = await session.run("What is for sale?")
print(reply.text)openjet project provisions the model into .openjet/models/ at build time, so your build bundles itBuilt for developers, on your machine
OpenJet is an open-source harness for developers who want to keep their code, prompts, and model weights on hardware they control — no hosted service in the loop, whether the model is answering you in a terminal or answering your users inside your app.
Setup profiles your machine, picks a model that fits it, configures the local runtime, and hands the same backend profile to the agent and the SDK.
OpenJet can read files, search the tree, apply edits, write new files, and run approved shell commands from the same CLI/TUI workflow.
Import the SDK, open an inference session, and call it. No server to stand up, no API key to manage, no request leaving the device.
Point openjet project at a use case, a target device, and a memory budget, and it recommends a configuration your machine can actually run.
The agent can propose terminal commands, but execution is approval-gated so state-changing work stays under your control.
Prompts, repo content, tool output, transcripts, and resume checkpoints stay on your machine instead of a hosted service.
Install
Run the install script in your terminal. OpenJet profiles your machine, configures the local runtime, and gives you both surfaces — the agent in your terminal and the SDK in your project.
pip install open-jetHead-to-head
Same prompt, same constraints: render an alien planet landing scene in a single HTML file. Each frame below is the model’s output, running live. See the Qwen3.8-27B local hardware and coding-agent guide.
Slipstream · opt-in
Local runs the loop. When a step genuinely needs a frontier model, Slipstream can hand that one step to the Codex subscription you already pay for — targeting a 20% Codex token share, around 5x more work per plan, measured live in your agent footer.
How the split works
Private agent evaluation
One controlled machine, one real repository, and a fixed-scope evaluation with evidence you can use to make a decision.
