Orientim ← Back to site

What Orientim does — and what it doesn't

Last updated: 26 July 2026

1. What it does

• Hard-stops an agent at your cap. Once the spend is reached, the next request is refused with HTTP 402.
• Caps the account, not just each agent. An optional ceiling across every agent, so per-agent caps cannot add up to a total you never agreed to.
• Resets on a schedule. Daily, weekly, monthly, or never for a one-off cap, so a cap is set once rather than cleared by hand each time it fills.
• Counts what actually costs money. Chat, streaming, and embeddings — including Anthropic's prompt-cache tokens, which are most of a Claude Code bill and are easy to miss.
• Warns before it stops. A desktop alert at 80% of the cap, and another when the agent is stopped.
• Keeps your provider key out of your tools. They talk to Orientim with a local key; the real one is added on the outbound call and never handed to them or to us.
• Runs entirely on your machine. No account, no cloud, no telemetry. One licence check at activation, and one more only if you deactivate a device yourself.

2. What it doesn't do

• It doesn't stop a request that's already in flight. A provider only reports cost after it answers, so the call that crosses your cap completes, along with anything already running. Your overshoot is the cost of those calls — one request for a single agent, more if your tool fans out in parallel.
• A request in progress when Orientim stops isn’t counted. If the app is closed, killed, or the machine loses power while a reply is still coming back, the provider bills that call and Orientim never sees the figures to record it. Your next total is short by that one request.
• It can't count what a provider doesn't report. Cost is metered from the usage figures in each reply. A few OpenAI-compatible servers — some self-hosted builds, some custom endpoints — return none, and for those the cap cannot move. Orientim marks those requests and says so on the dashboard rather than showing you a limit that isn’t holding.
• It doesn't see traffic that doesn't pass through it. If a tool talks to the provider directly, or Orientim isn't running, nothing is counted. (Agents pointed at Orientim simply fail while it's closed — they don't quietly bypass it.)
• It can't cap a flat-rate subscription. Orientim meters a provider API key billed per token. If you use Claude Code on a Claude Pro, Max or Team plan — signing in with /login rather than pasting an sk-ant- key — then your tokens are covered by the monthly plan and there is no per-request dollar amount for anything to count or stop. The same goes for Codex on a ChatGPT plan. Orientim has nothing to meter there, and pointing Claude Code at an API key while you are signed in will make it complain about conflicting credentials rather than fix anything.
  
  How to tell: if you have never pasted a provider API key into anything, and you have never received a usage bill from Anthropic or OpenAI separate from your subscription, you are on a subscription and Orientim is not for you yet. Switching to API billing to gain a cap usually costs a heavy user more than the flat plan does.
• It doesn't work with Cursor or GitHub Copilot. Those route through their own servers, so a local proxy never sees the requests. Claude Code, Codex CLI, Continue, Cline, Aider and any OpenAI- or Anthropic-compatible SDK do work.
• It doesn't know the price of a model it has never heard of. A model that matches nothing in the table is billed at $20 per million in, $80 per million out — set at the top of what any current model charges, not the middle, so an unknown model is over-counted rather than under-counted. That means your cap trips early with a message naming the model, instead of late with a bill. Every such request is flagged in the dashboard as a guess, and you set the real price in one field on the Pricing page. Until you do, the cap for that model is an estimate on the cautious side.
• It doesn't encrypt the key file. Your provider key sits in a file in your Windows user profile, like any other credential file on your machine. Anyone with access to your account can read it.
• It isn't DRM or a security boundary. It's a safety net for your own agents, not protection against someone who controls your computer.
• It doesn't manage your provider account. Your keys, your billing, your relationship with OpenAI or Anthropic — all still yours.

3. How accurate are the numbers?

OpenRouter is exact — it reports the real cost of every call and Orientim bills with that figure. Everything else is calculated from token counts and a price table shipped with the app, which you can edit at any time in USD per 1M tokens, the same unit providers publish.

That means Orientim's total will track your provider's invoice closely but will not match it to the cent — rounding, promotional pricing and provider-side billing rules are outside what a local proxy can see. It is a meter, not a reconciliation of your invoice.

4. What we support

Email support@orientim.com and we will help with:

• Installing, activating, or moving your licence to another machine
• Connecting a tool or SDK to Orientim
• Anything Orientim itself does wrong

What support does not cover:

• Your provider's invoice, quotas or rate limits
• What your agent chooses to do, or how many tokens it decides to spend
• Spend that happened while Orientim wasn't running or wasn't in the path

5. Requirements

• Windows 10 or 11 (64-bit). No macOS or Linux build yet.
• Ports 8000 and 8501, or the next free pair — Orientim moves up if either is taken.

6. Trial, licence and updates

• Free until Orientim has metered $20 of your own spend — the whole product, no card, no clock. The meter only advances while Orientim is working, so it can’t run out while the app sits idle. At $20 it stops passing requests until a licence key is entered; it never carries on with the caps switched off. Nothing you set up is lost.
• One licence activates up to three machines; free a slot yourself with Deactivate this device.
• There are no updates. You buy this version as it stands and keep it. No roadmap, no version 2. The price table can go stale as providers change rates — unknown models bill at a high fallback rate and are flagged, and you correct any price yourself on the Pricing page, permanently.

Operated by Qazim Memeti, North Macedonia.