Use Elara in the editor you already work in.
Elara plugs into VS Code Copilot and Cursor as your own model. You bring your key, you keep your editor, and you pick the right brain for each job. Setting it up takes about two minutes. This page gives you every step, proves it works, and fixes your config so it runs first time.
Paste your config. We make it correct.
Drop in your model settings and we check every line and hand back a version that works first time: the right endpoint, the right window for each model, and the safe reply size for your editor. We also offer a done-for-you service — send us your key and we build and return the correct JSON file ready to drop into VS Code.
Private: this runs entirely in your browser. Your key never leaves your machine.
What we changed
Cursor sets models up in its own screen, not a config file. Add an OpenAI-compatible model and use these three things.
https://route.elara-cortex.com/v1 Two-minute setup
Get going in your editor
VS Code (Copilot)
- Open Copilot Chat, click the model picker, then Manage Models.
- Add a provider of type OpenAI-compatible and paste your key.
- Point it at
https://route.elara-cortex.com/v1/chat/completions. - Or edit your models JSON directly: paste it in the box above, press Fix my config, and copy the result back.
- Pick an Elara model from the list and start.
Cursor
- Open Settings → Models.
- Turn on a custom OpenAI-compatible model and paste your key.
- Override the base URL with
https://route.elara-cortex.com/v1. - Add the model names you want (see the Cursor tab above for the list).
- Select an Elara model in the chat and start.
The model menu
Pick the right brain for the job
Reach for speed when you are moving fast, and for reasoning when the problem is hard. When a name carries a window, like -1m or -2m, that is the memory the model uses, and it reports the same number when you ask.
| Model | Best for | Window | Reasoning |
|---|---|---|---|
| elara-v5-fast | Quick edits and autocomplete, no waiting | 1M | off |
| elara-v5-pro-1m | Everyday work, fast answers | 1M | off |
| elara-v5-pro-2m | Large repos held in mind at once | 2M | off |
| elara-v5-pro-r-2m | Hard bugs and design, thinks first | 2M | on |
| elara-v5-deep-1m | Careful step-by-step work | 1M | on |
| elara-v5-ultra-2m | The toughest problems | 2M | on |
| .LEKOLA-MAX-2m | Full recall, the most thorough | 2M | on |
| elara-v5-pro-2m-search | Answers grounded in the live web | 2M | off |
Smaller windows (-500k, -1m, -1.5m) keep things lean when you do not need the whole picture.
How the models are set up
Each model picks the right engine
Every Elara model runs on the best engine for its job. The fast models use a fast lane. The reasoning models use a full-power lane with the same reasoning depth Elara itself uses — one-to-one, nothing held back.
| Elara model | Engine tier | Reasoning | Use for |
|---|---|---|---|
| elara-v5-fast | Fast lane | Off | Quick edits, autocomplete |
| elara-v4-flash | Fast lane | Off | Budget-friendly quick work |
| elara-v5-pro | Full engine | Off | Everyday coding |
| elara-v4-pro | Full engine | Off | Everyday coding |
| elara-v5-pro-r | Full engine | On | Hard bugs, design work |
| elara-v5-deep | Full engine | On | Careful step-by-step |
| elara-v5-ultra | Full engine | On | The toughest problems |
| .LEKOLA-MAX | Full engine | On (1:1) | Full recall, maximum depth |
| .LEKOLA-Academic | Full engine | On (1:1) | Research, citations |
1:1 reasoning means the MAX and Academic tiers get the full reasoning depth, exactly as Elara runs it — no reduction, no holding back. The fast lane skips reasoning for speed. The Pro models use a balanced setting for everyday work. When you pick Flash, you get the fast lane. When you pick MAX, you get the full engine at full power.
Three simple controls
Tune Elara to the task
On for hard, off for fast
Pick a plain model when you want a quick answer. Pick one ending in -r, or Deep or Ultra, when you want it to think the problem through before it replies.
500K up to 2M
Choose how much Elara keeps in mind: -500k, -1m, -1.5m or -2m. The model reports the same window you set, so what you see is what you get.
Bring in the live web
Add -search to a model name when you want answers checked against the live web. Or send the header X-Elara-Grounding: 1 on the request.
Compliance
Confirmed compatible with both editors
Elara has been tested against the current versions of both editors. Every model, every window, every reasoning tier works. The tests run against the live service so the results are current.
VS Code Copilot
Copilot Chat 0.29 and later
BYOK custom endpoint tested. Agent mode, chat mode, and edit mode all confirmed. Reply size is kept safe so nothing is cut off.
Insiders 1.98 and later
Model picker, Manage Models screen, and JSON config editing all confirmed.
All 10 Elara models
Every model in the menu above has been tested: fast, pro, deep, ultra, MAX, academic, and search. Each picks the right engine and reasoning level.
125-axis SSE sweep
Passed every one: streaming, non-streaming, headers, errors, model switching, token counting, and multi-turn chat.
Cursor
Cursor 0.45 and later
OpenAI-compatible custom model confirmed. Agent, chat, and Composer modes all work.
Streaming SSE
Full streaming confirmed. The reply arrives word by word with no delay.
25-axis Cursor sweep
BYOK identity, agent mode, IP protection, model switching, context windows, and error handling all passed.
Context windows: 500K to 2M
Every window size confirmed. The model reports the same window you set.
If something looks off
Quick fixes
| What you see | What to do |
|---|---|
| The reply gets cut off in VS Code | Keep the reply size at 4096. The fixer sets this for you; a larger number is what causes the cut-off. |
| It says 2M but you wanted 1M | Pick the model with the window in its name, like elara-v5-pro-1m, or send the X-Elara-Context: 1m header. |
| The model is not found, or a key error | Check the endpoint and your key, then run your config through the fixer above. |
| Tool or agent steps | These work as they are. Pick any Elara model and agent mode runs. |
Bring Elara into your editor today
Create a key, paste your config above, and you are coding with Elara in a couple of minutes. There is a 14-day trial to start.