DEDUPED Hot AI repos. Real use cases. What people build. Get the Monday 5

How to run a coding agent fully offline on your own GPU with Open Jet

Install OpenJet with pipx, let setup pick a model for your hardware, and start coding offline.

Short answer: Run pipx install open-jet; then run openjet setup and accept the recommended model (or openjet setup --yes); on an RTX 3090, 4090 or 5090, choose Qwen3.8 27B Q4_K_M MTP so it builds the megakernel. All steps.

Who it is forA developer who wants a coding agent with no API bills and no code leaving the machine.

The problemCloud coding agents cost per token and send your repo to a provider.

The outcomeA local coding agent running a model sized to your VRAM, working offline after the download.

What you need first

  • Linux with an NVIDIA RTX 30, 40 or 50 GPU with 24 GB+ and CUDA 12+ for the megakernel; other hardware uses llama.cpp
  • Recommended: 14 GB+ VRAM or Apple silicon with 24 GB+; CPU-only works
  • pipx

Steps

  1. Run pipx install open-jet
  2. Run openjet setup and accept the recommended model (or openjet setup --yes)
  3. On an RTX 3090, 4090 or 5090, choose Qwen3.8 27B Q4_K_M MTP so it builds the megakernel
  4. Start the agent with openjet in your repo
  5. If setup prints Skipping the Qwen3.8-27B megakernel, read the reason; it falls back to llama-server

Prompt for your agent

Install L-Forster/open-jet with pipx, run openjet setup for this machine's GPU, then use openjet in <repo> to <task>. Tell me which model and engine it chose and why.

Caveat. Setup downloads about 17 GB for Qwen3.8-27B and compiles llama.cpp; budget the disk space and time.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for L-Forster/open-jet and "Run a coding agent fully offline on your own GPU"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer