Speak naturally
Use multiple languages in one task. Change language mid-sentence and DJL keeps the same working context.
DJL FOR DESKTOP
A capable agent,on your desktop.
Describe the outcome. DJL plans the work, uses the right tools, and pauses when your judgment matters.
macOS 13+ · Windows available
New ChatHand off
What should we work on?
Product
One window. The whole loop.
Plan, tools, diff, review — the entire run is visible in a single desktop window, and nothing is applied until you approve it.
Describe the outcome. Watch the work. Approve the change.
Capabilities
Your work. Any model.
Work in the language that feels natural. Switch between API and local models in one click, move from light to dark, and keep the task intact.
Switch models, keep context
Move to a different model in one click. Your task, files, and instructions stay exactly where they are.
API to fully local
Reach hosted models when you need them, or run privately through Ollama and LM Studio on your own machine.
Light or dark
Follow your system or choose the appearance that keeps long work sessions comfortable.
Local AI
Actually local.
Pick a model sized to your hardware and DJL runs it on-device — planning, tool calls and edits included.
- Runs with the network cable out
- Curated models from 1.7B to 30B
- Memory requirements stated up front
Local vs API
Why an API model is still the workhorse
Local models earn their place — but bigger tasks want a bigger brain. Here is the honest comparison, and how DJL lets you use both.
API model
- Frontier-level reasoning — plans further ahead and untangles harder bugs than anything that fits in local memory
- Long context: holds far more of your codebase in one task
- Full speed on any machine — inference runs in the datacenter, not on your fans
- Dependable tool-driving, from the first call to the last diff
Routed over a secure connection — and only when you choose.
Local model
- Code never leaves the device
- Works offline; nothing to pay per token
- Capability tops out at 30B — frontier models are a league above
- Wants 4–32 GB of memory; the best local models need the biggest machines
- The smallest models chat but can't drive tools
The DJL answer: run both. Keep private, everyday work local — and when the task is bigger than the machine, route it to the API. Every change is review-gated either way, and your keys stay home.