FAQ
The questions that arrive most often, answered in one paragraph each.
Yes — the app, all of it, licensed MIT, with no tier above it. There is nothing on this site or in the binary that takes payment. Your provider accounts are the only thing that costs money, and you pay those providers directly, exactly as you do today. Connect only free tiers and the whole stack runs at $0.
Because there is no server in the middle to pay for. ShareQuota runs on your machine, holds its state in a SQLite file in your data directory, and talks only to the providers you connected. There is no infrastructure bill, so there is nothing to recover.
The MIT licence ships in the archive beside the binary, and every release is published with a SHA-256 checksum file so you can verify exactly what you downloaded. The repository itself is not published yet.
Into a local SQLite database in your data directory, created with 0700 permissions. They are never transmitted anywhere except to the provider they belong to. Clients on your machine only ever see the internal sq- key, so a leaked client config leaks nothing that can be billed.
There is nobody in a position to. Requests go from your machine straight to the provider — there is no relay, no account, and no telemetry. Nothing about a request leaves your machine except the request itself, addressed to the provider you chose.
Yes, and switching them over is one click in the dashboard. The original config file is copied to a backup directory before the first write, and disabling restores it exactly — so trying it costs nothing you cannot undo.
Because these five are the ones with subscriptions worth pooling, and because breadth without tests is a list rather than a feature. The binary is 157,679 lines of Go under 3,015 race-enabled tests, and the translator layer is held to golden fixtures generated from a reference implementation. If you need a model none of the five carry, OpenRouter is one of the five, and it carries most of the rest.
OpenRouter is a hosted service that sells API credits: you pay them per token, and you cannot put a Claude Pro or ChatGPT Plus subscription through it. ShareQuota runs on your machine and signs in to the subscription you already pay for. If what you want is 400 models with nothing to install, OpenRouter is the better choice and this page will not pretend otherwise.
No. There is no sign-up, no login, and no account — on the site or in the binary. The only credential involved is the sq- key the binary generates on your machine on first run.
No. ShareQuota has no server between you and the provider, so there is nothing it could meter or bill. It signs in to your own accounts, on your machine, and routes your requests to them. Nobody else’s quota is involved, and yours is never sold.
You connect your own accounts, on your own machine, and requests reach each provider through its own official endpoints — nothing is scraped and nothing impersonates a browser. Provider terms differ on tooling around subscription access, and they change; read yours, and weigh it the same way you would any local tool that automates an account you own.
The router moves the request to another connection that can serve the same model, and puts the exhausted one in a cooldown with exponential backoff rather than hammering it.
Yes. It is a single static binary with no runtime and three direct dependencies, and the repository ships a systemd unit. The management API is bound to localhost by default, so put it behind your own auth before you expose it.
One directory: ~/.sharequota, holding the SQLite database and the backups of any CLI config it changed. Disable the CLI integrations to restore those configs, then delete the directory and the binary. Nothing else is written anywhere.
An AI router — also called an LLM proxy or LLM gateway — is a program that sits between your tools and several model providers, exposes one endpoint, and decides per request which provider answers, usually with failover when one refuses. ShareQuota is one, with a narrow focus: the five providers whose subscriptions are worth pooling, running as a single binary on your own machine.
For one person on one machine, yes: it does the same core job — one local endpoint, subscription login, auto-failover across your own accounts — as one Go binary with no runtime. It is not a replacement if you want hundreds of providers, token compression, MCP or a gateway a whole team shares; both of those projects do those things and this one does not. The comparison pages go row by row.
Yes, through the OpenAI-compatible endpoint. Cline and Copilot are pointed at localhost from their own settings. Cursor sends its requests from Cursor’s own servers, so the endpoint has to be published through the built-in tunnel first. Each tool has its own setup page under Integrations.
The real channels, with response expectations.AI router, LLM proxy, failover, dialect, quota — the terms, defined.