# proxium docs

> proxium is an OpenAI-compatible LLM gateway. A client sends one base URL and one virtual key. proxium routes the call, enforces the budget of the project and records the real cost.

## Get started

- [Get started with Proxium](https://proxium.tech/docs/index.md): Proxium is an LLM gateway for the OpenAI and Anthropic APIs. Your apps send model calls with one virtual key, and Proxium routes them to your providers.
- [Quickstart](https://proxium.tech/docs/quickstart.md): Add a vendor, create an endpoint, pick a model and send your first request through Proxium.

## Connect your app

- [Use the OpenAI SDK](https://proxium.tech/docs/guides/openai-sdk.md): Connect an app that uses the OpenAI SDK for Python or Node, or plain HTTP, to Proxium. Send requests, stream replies and handle errors.
- [Use the Anthropic SDK](https://proxium.tech/docs/guides/anthropic-sdk.md): Point the Anthropic SDK for Python or Node, or curl, at Proxium. Send Messages API calls, stream them and read errors.
- [Use the Responses, embeddings, rerank and moderations APIs](https://proxium.tech/docs/guides/other-apis.md): Send Responses API, embeddings, rerank and moderations calls through Proxium, and know which vendor answers each one.
- [Generate images, speech and video](https://proxium.tech/docs/guides/images-speech-video.md): Send image, speech and video calls through Proxium, poll a video until it is ready, and know what these routes do not do.
- [Connect an MCP client](https://proxium.tech/docs/guides/mcp.md): Add the Proxium MCP server to Claude Code, Claude Desktop, Codex, Cursor, VS Code, Gemini CLI or Zed, so that your agents read and save the memory of your project.

## Set up your project

- [Create endpoints and keys](https://proxium.tech/docs/guides/endpoints.md): Make an endpoint, a base URL and a key with its own routing, for your apps. Know when to make a second endpoint, and when a plain key is enough.
- [Add vendors and your own keys](https://proxium.tech/docs/guides/vendors.md): Add a model vendor such as OpenAI or Groq with your own key, add a server that you run yourself, and change the models of a vendor.
- [Invite your team](https://proxium.tech/docs/guides/team.md): Add a colleague to a project as a member or an owner, know what each role can do, and remove a person.

## Run in production

- [Route requests to models](https://proxium.tech/docs/guides/routing.md): Send a model id or a tier name, set the models of a tier, give one app its own models, and know what Proxium does when a model fails.
- [Set budgets and limits](https://proxium.tech/docs/guides/budgets-and-limits.md): Limit how many calls an app or a key makes and how much it spends in a day, handle the 429 at a limit, and check a key before a call.
- [Track spend](https://proxium.tech/docs/guides/track-spend.md): See what each call cost, split the spend by application, key, model and task, and read the spend of the month from the API.
- [Monitor your project](https://proxium.tech/docs/guides/monitor.md): Watch the spend, the failures and the vendors of your project in the console and from the API, and know which monitoring Proxium does not offer yet.
- [Debug a failed call](https://proxium.tech/docs/guides/debug-a-failed-call.md): Find why a call failed. Tell a refusal of Proxium from a failure of the vendor, read the attempts of the call on the Requests screen, and keep the prompts of failed calls.
- [Lower your costs](https://proxium.tech/docs/guides/lower-costs.md): Read the findings of the Improve screen, move traffic to a cheaper model, let the cache answer repeated calls, and fix a vendor that fails often.
- [Use project memory](https://proxium.tech/docs/guides/memory.md): Turn on project memory, choose what each call does with it, give your agents the memories, and erase one end user.

## Concepts

- [The request flow](https://proxium.tech/docs/concepts/request-flow.md): The steps of a chat call through Proxium, from the key check to the cost record, and the reason for their order.
- [Automatic routing](https://proxium.tech/docs/concepts/automatic-routing.md): How Proxium can choose a tier for each call that sends model auto, with a classifier model, and what model auto does on proxium.tech today.
- [The response cache](https://proxium.tech/docs/concepts/caching.md): How Proxium caches chat and Messages answers per project, what a cache hit costs, and which calls the cache never serves.
- [Sensitive data scanning](https://proxium.tech/docs/concepts/sensitive-data.md): How Proxium can find card numbers, IBANs, email addresses, phone numbers and credentials in a request, and redact or block them before the request leaves.
- [Data handling](https://proxium.tech/docs/concepts/data-handling.md): What Proxium stores about each call, how long it keeps it, and how a project exports or erases its data.
- [What Proxium does not do](https://proxium.tech/docs/concepts/not-supported.md): The gaps a caller of proxium.tech meets, such as no speech to text, no export of call records and no spend alerts, and what to do instead.

## Providers

- [Providers](https://proxium.tech/docs/providers.md): The vendors that proxium can call with your own key, and how to add a vendor that is not in the list.
- [OpenAI](https://proxium.tech/docs/providers/openai.md): Add OpenAI to a proxium project with your own key, and call its chat models through proxium.
- [Anthropic](https://proxium.tech/docs/providers/anthropic.md): Add Anthropic to a proxium project with your own key, and call its chat models through proxium.
- [Groq](https://proxium.tech/docs/providers/groq.md): Add Groq to a proxium project with your own key, and call its chat models through proxium.
- [OpenRouter](https://proxium.tech/docs/providers/openrouter.md): Add OpenRouter to a proxium project with your own key, and call its chat models through proxium.
- [Mistral](https://proxium.tech/docs/providers/mistral.md): Add Mistral to a proxium project with your own key, and call its chat models through proxium.
- [DeepSeek](https://proxium.tech/docs/providers/deepseek.md): Add DeepSeek to a proxium project with your own key, and call its chat models through proxium.
- [Together AI](https://proxium.tech/docs/providers/together.md): Add Together AI to a proxium project with your own key, and call its chat models through proxium.
- [Google Gemini](https://proxium.tech/docs/providers/gemini.md): Add Google Gemini to a proxium project with your own key, and call its chat models through proxium.
- [OpenAI (embeddings)](https://proxium.tech/docs/providers/openai-embed.md): Add OpenAI (embeddings) to a proxium project with your own key, and call its embeddings models through proxium.

## Reference

- [Console screens](https://proxium.tech/docs/reference/console.md): Each screen of the Proxium console, what it shows, what you can change there, and the page that explains it.
- [Request headers](https://proxium.tech/docs/reference/headers.md): Every HTTP header that a caller can send to Proxium, and every response header that Proxium sets, with values, defaults and effects.
- [Errors](https://proxium.tech/docs/reference/errors.md): The error shapes of Proxium, how to tell a refusal of Proxium from a failure of the vendor, and every error code with its status, cause, fix and retry rule.
- [Limits](https://proxium.tech/docs/reference/limits.md): Every size, time, retry, spend and name limit that a caller of proxium.tech can reach, with its value and what happens at the limit.

## Optional

- [OpenAPI document](https://proxium.tech/docs/openapi.json): Every endpoint of the API reference, as OpenAPI JSON.