Orientim
Setup Guide Contact Us

A dollar limit your agent cannot cross · 01 / 09

You set the cap. Your agent can't cross it.

A local desktop app that hard-stops your AI coding agents at a dollar limit — not a warning after the money's gone. Caps per agent and across the account, budgets that reset themselves, and per-token pricing accurate to the cent it stops on.

Free until it has metered $20 of your own spend No credit card required $39.99 lifetime key once
agent · claude-code-dev● metering
$0.00 / $50.00 cap
↕ scroll down — watch spend climb toward the cap

While you were asleep · 02 / 09

One stuck loop, and $100 is gone by morning.

You leave a coding agent running while you sleep. It hits a loop. The provider's dashboard warns you — hours later, once the money's already gone. Nothing, anywhere, actually stops it.

Provider dashboards report spend after the fact. Nothing in that path refuses a request.

overnight · unattendedno limit
$0.00
11pm3am7am
nothing stopped it.

What you connect · 03 / 09

Works with the stack you already run.

One local link between your tools and every provider you run.

OpenAI Anthropic Gemini Groq DeepSeek OpenRouter
Works with Claude Code, Codex CLI, Continue, Cline, Aider — with your own code via LangChain, LlamaIndex, AutoGen, CrewAI — and with any OpenAI-compatible endpoint you point it at: xAI, Mistral, Together, a local vLLM. If you can set a base URL and the request leaves your machine, it counts, including providers released after this build.
Chat, streaming and embeddings all bill against the same cap — so a RAG index or an agent's memory can't quietly spend around it.

Your key, protected · 04 / 09

Every agent runs. None of them holds your key.

# point your tool at Orientim, use the local key:
BASE_URL = http://127.0.0.1:8000/proxy/<agent>
API_KEY  = orientim-secure-key
tool sends orientim-secure-key Orientim swaps in sk-…real official API

An API key is unlimited authority — whoever holds it can spend without a ceiling. Orientim keeps that authority and hands your agents a key with a budget attached. You haven't added a meter to watch; you've issued a credential they can't overspend.

How it works · 05 / 09

Forty seconds. Watch it refuse.

A quick tour of the dashboard — everything you'll actually use, in the order you'll use it.

  • Add an agent and paste its provider key
  • Set a hard dollar limit
  • Track spend and exact cost, per provider
  • See an agent hit its cap and turn red
  • Read each request — model, tokens, exact cost

The panel · 06 / 09

One panel. Every agent, every provider, live.

Independent limits per key. When one hits its ceiling it turns red and stops — everything else keeps running.

🟠 Claude Code Devanthropic
$35.00 / $35.00
● Limit reached
🟢 Production GPT-5.6openai
$45.20 / $50.00
● Active · 90%
🔵 Gemini Experimentsgemini
$1.24 / $10.00
● Active · 12%
🔶 OpenRouter Poolopenrouter
$12.80 / $30.00
● exact cost from provider

Every call is metered from each provider's own rates. If a provider changes its pricing, you edit it yourself in the dashboard — changes apply to the very next call.

What’s inside · 07 / 09

One licence. The whole instrument.

Not a meter with a limit bolted on — four things that have to be right before a hard stop means anything.

Does it stop?control
  • A hard cap per agent. Cross it and the next request is refused, not logged.
  • An account ceiling above them all — when the total lands, every agent stops.
  • Budgets that reset themselves: daily, weekly, monthly, or one-off forever.
  • A desktop alert at 80%, and another the moment it stops — each once per crossing, not on every call.
  • Stop all agents in one press, and pause or resume any one of them.
Is the number right?accuracy
  • Anthropic prompt caching is priced properly — cache reads at 0.10×, writes at 1.25×. Claude Code relies heavily on the cache, and pricing those tokens as ordinary input makes a cached agent look ten times more expensive than it is. OpenAI's cached input is priced at its own discount; providers that report no cache split bill those tokens at the full input rate.
  • Embeddings billed at embedding rates for OpenAI, Gemini and Voyage models, not at a chat rate that would overcharge them enormously and trip your cap on the first index. An embedding model we don’t list is still priced as one, and flagged so you can correct it.
  • OpenRouter is exact — it reports real cost, so we use its figure rather than an estimate.
  • Every rate is yours to edit, and a model too new to be listed can be priced by hand.
Will it bother me?daily use
  • Test connection sends one real token through the whole path, so a mistyped key fails while you’re still looking at it.
  • Pick your client and it writes the setup line for you — the id is never retyped.
  • Logs keep the input/output split, the stop reason and which price was used, so a surprising number can be traced.
  • Offers to start with Windows during install.
  • Installs without admin rights, and your agents, keys and history survive an uninstall.
What stays mine?trust
  • Runs entirely on your machine. No account, no cloud, no telemetry.
  • Your provider key is masked everywhere and injected onto one request only — the one going to the provider it already belongs to.
  • Your tools never hold the key. They hold a local one that works nowhere else.
  • One licence check when you activate — and one more only if you release the machine yourself. Nothing else, ever.

Free until it has metered $20 of your own spend then $39.99 once · no subscription

Questions · 08 / 09

The questions worth asking before you install it.

QIs my API key safe?
It stays in a file on your own machine and is added only on the outbound HTTPS call to the provider it already belongs to. We never receive it, and your tools never hold it — they talk to Orientim with a local key instead.
QWindows warns me when I run it. Why?
The installer is not yet code-signed, so Windows SmartScreen does not recognise the publisher. Click More info, then Run anyway. To confirm the file is the one published here, check its SHA-256: fe4d1a4228f3817a1fb559fa809425aa89b6db362596229bb1dd8f26b8b75416.
QWill I know when an agent is stopped?
Twice. A desktop alert at 80% of the cap, while there's still room to act — and another the moment the agent is stopped. Each fires once, not on every request. Raise the cap and it resumes.
QDoes it phone home?
One license check when you activate, and one more only if you click “Deactivate this device” to free the slot for another machine. Nothing else, ever — no account, no telemetry, no update pings. Every other packet it sends is your own request going to your own AI provider. Verify with Wireshark.
QDoes it work with Cursor?
No. Cursor and GitHub Copilot route requests through their own servers, so a local proxy never sees the traffic. It works with Claude Code, Codex CLI, Continue, Cline, Aider, and anything that calls the API directly from your machine.
QI’m on a Claude Pro or Max plan. Does it help?
No. Those plans charge one flat monthly fee, so your tokens have no individual price and there is no dollar figure for anything to count or stop. Orientim meters a provider API key billed per token — an sk-ant- or sk- key you paste in yourself. Quick test: if you sign into Claude Code with /login and have never had a usage bill separate from your subscription, this is not for you yet. Same for Codex on a ChatGPT plan. Switching to API billing just to gain a cap usually costs a heavy user more than the flat plan does.
QWhat are its limits?
It stops the next request, not mid-request — a provider only reports cost after it answers, so the call that crosses your cap finishes, and so does anything running beside it. The overshoot is exactly that: whatever was already in flight. One agent working normally has one request in the air, so you land within pennies of your number. Fire twenty calls at once and you carry twenty. It is a spending limit, not an access control.
QCan I try it before paying?
Yes — the whole product, until it has metered $20 of your own API spend. That's traffic you were sending anyway, not money you give us. No card, no account, no clock: the meter only moves when Orientim is working, so it can't expire while you're busy. When it's reached, requests stop — Orientim never keeps forwarding with the caps switched off.
QCan I cap everything at once, not just per agent?
Yes. An account cap sits above every agent — when their combined spend reaches it, all of them stop. That is the number that lands on your bill. Each agent still has its own cap underneath it.

The hard stop · 09 / 09

Draw the line before your agent finds the edge.

Install it in a minute, decide later. A single unattended loop can run to three figures overnight.

Free to start / then $39.99 once
agent · claude-code-dev● at the cap
$50.00 / $50.00
HTTP 402 — BLOCKED. No further spend until you raise the limit.
$0.00 / $50.00
SPEND
← → keys · scroll · swipe