Start free. Decide with your own figures.

Right now there is only the free tier, and it is free because nothing is for sale yet rather than because it is crippled. Paid tiers are planned, and their prices are below so you can see where this is going.

Free

5k saved per prompt

50 retrievals a month · no account required

  • 4K context window
  • Single-file tasks
  • Local search and context packets
  • The measurement dashboard
  • Community support on GitHub
Install

Pro

15k saved per prompt

Planned: $20 a month, or $210 a year

  • 6K context window
  • Multi-file tasks
  • Local drafting and delegation
  • Multi-agent task board
  • Email support, next business day
Not yet available

Max

No ceiling on what one prompt saves

Planned: $100 a seat monthly, or $1,000 a seat yearly

  • 12K context window
  • Everything in Pro
  • Shared task boards
  • Audit logs for compliance review
  • Air-gapped deployment guide
  • Priority email support, next business day
Not yet available

Why the price is in tokens, not percentages.

Each tier caps how much a single retrieval may keep out of your agent's context. Not a monthly budget — a ceiling on one answer.

We can't sell you a percentage, because the honest one depends on your codebase, your queries, and your model. A number that's true on a monorepo is a lie on a small project. A token ceiling is something we can actually promise, and something you can check in the dashboard.

Measured on this project's own repository, an unbounded retrieval saves about 29,000 tokens. So Free returns a real but bounded packet, Pro clears most ordinary calls, and Max is for repositories where the honest answer is simply bigger than any ceiling we'd pick.

Task size still differs too: Free covers single-file work, which is also what the local model can carry on its own once your usage window is spent.

None of the paid tiers exist yet. There is no checkout, no card form and no way to pay us even if you wanted to — the prices are published early so nobody installs the free tier under a wrong impression of where it leads. When paid tiers arrive, the free tier keeps everything it has today.

Install

Three commands, then open the sidebar.

ollama pull qwen3:8b
code --install-extension latorium
# open the Latorium panel and run Check setup

Latorium doesn't ship a model — pull whichever one you want to run. If Ollama isn't reachable the panel says so plainly instead of failing quietly.

4K context · Free
5.97 GB · 8 GB VRAM
6K context · Pro
6.27 GB · 8 GB VRAM
8K context
6.57 GB · 10 GB VRAM
12K context · Max
7.20 GB · 12 GB VRAM
System memory
16 GB
VS Code
1.95 or later

The first figure is the measured footprint of qwen3:8b at that window; the second is the card we would want, allowing headroom for your display. A GPU is optional — everything runs on CPU, just slower.

What matters is whether the whole model fits. On a 6 GB card the 8B spills about a third of itself to system memory and runs at roughly a quarter the speed, and a larger context window makes that worse rather than better: measured on one, GPU residency fell from 70% at 4K to 57% at 12K. Pick the window your card can actually hold. Air-gapped install guide