Who it is forA developer who wants frontier-model planning but hits plan limits on long coding sessions.
The problemFrontier models burn through quotas when they read the whole repo and run every test themselves.
The outcomeA frontier model that plans and reviews while the local model edits files and runs tests, using about 20% of the tokens.
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
- Install and set up OpenJet with
pipx install open-jetandopenjet setup - Inside OpenJet run /connect openai-codex (or /connect openrouter)
- Switch with /mode to Slipstream
- Give it a task; the local model reads the repo, edits files and runs tests
- Review the frontier model's review of the local model's conclusions
Prompt for your agent
Set up L-Forster/open-jet in Slipstream mode with my Codex subscription as planner and the local model as worker, then do <task> in <repo> and report tokens used by the planner.
Caveat. This card is written from the L-Forster/open-jet README, not from a test run by Deduped. Before you start, make sure you have Linux with an NVIDIA RTX 30, 40 or 50 GPU with 24 GB+ and CUDA 12+ for the megakernel. Check the repo for newer instructions before you rely on these steps.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for L-Forster/open-jet and "Let Codex plan while a local model does the edits (Slipstream)"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer