--- title: "Getting Started" id: "1932" type: "page" slug: "getting-started" published_at: "2026-09-30T22:28:02+00:00" modified_at: "2026-10-01T00:11:02+00:00" url: "https://xedant.com/agents/proxy/docs/getting-started" markdown_url: "https://xedant.com/agents/proxy/docs/getting-started.md" excerpt: "Proxy Agent is one program on your server through which all your programs reach AI…" --- # Getting Started [https://xedant.com/agents/proxy/docs/getting-started.md](https://xedant.com/agents/proxy/docs/getting-started.md) Proxy Agent is one program on your server through which all your programs reach AI models. Instead of getting keys from a dozen vendors and spreading them across computers, you keep the vendor keys in one place and issue your own access keys to people and contractors — with a spending limit and a list of allowed models. Along the way the product keeps a request log, counts spending in money, can hide secrets out of requests and route ordinary — non-AI — traffic through the same server. ## The main idea Models and vendors keep multiplying, and the mess grows with them. Vendor keys spread across employee and contractor computers, nobody remembers which is where, and the bill at the end of the month arrives as one confusing number. Proxy Agent gathers all of it into one point: programs talk not to ten vendors but to one address — your server. What this gives you in practice: - **Vendor keys stop spreading across computers** — they live in one place, and every person gets their own access key. - **Spending shows in money, not in tokens** — you set the model prices, and the product computes the cost of every request. - **A spending limit can be set in advance** — per day, week, month or for the whole lifetime; exhausted — access closed. - **Every request leaves a record** — who, which model, how much it cost, through which channel it went. - **A key can be revoked instantly** — a contractor’s project ends, and their access closes with one motion. ## Who it fits - **A small team** — several people use AI from different programs, and you want to see how much it costs and who exactly is spending. - **A company with contractors** — external workers need access to the models, but not to your vendor keys. - **Anyone with several vendors** — one program works with models from different vendors, and switching between them by hand is inconvenient. - **Anyone who cares about secrets** — keys, passwords and document numbers can be removed from requests in advance, so they never reach the model. - **A small studio or department** — you also want your own proxy for ordinary traffic: browsers, utility programs, bots. ## How it differs from “just a vendor key” A vendor key is a pass, but with no accounting: who uses it, how much they spend and on what — you do not know. Proxy Agent stands between the programs and the vendor and adds order to that pass: its own access keys for people, spending limits, a log and analytics. A separate word about the similar name. The site has a [“Proxy Server”](/agents/agent/docs/proxy) page — that is a **different thing**: the proxy built into Xedant Agent, which the agent itself needs for working with models. Proxy Agent is a standalone product: all your programs use it, not just one agent. ## What the product consists of A few words that keep coming up below. Almost all of them are section names in the interface itself. - **Dashboard** — a summary: how many models and keys, request volume, error share, spending and proxy service traffic. - **Access Keys** — the access keys: who has which, with what limits and for which models. - **Models** — records of where and how to reach a model, and what a million of its “portions of text” (tokens) costs. - **HTTP proxies** — channels out to the internet: your own proxy, SOCKS5, a VPN or a Happ subscription. Plainly put — where your traffic exits. - **Request Logs** — records of every request to a model, and separately of the proxy service’s traffic. - **Analytics** — the log turned into numbers: volume, latency, breakdowns by models, keys and errors, spending over time. - **Anonymiser** — the rules by which secrets are replaced with conventional placeholders before going to the model, and put back in the reply. - **Settings** — language and font size, the secret-hiding master switch and the leak report retention. At the bottom of the window — the license line; clicking it opens its details. If you connect an external Xedant Agent, a chat with the AI agent appears right inside the interface — you can ask what the log and analytics show, and get help with configuration. One more thing worth knowing from the very start: Proxy Agent checks the license. A valid key is needed for traffic to flow through the server, while the interface, settings, logs and management over the API stay open without it. The license state is visible in the line at the bottom of the “Settings” window; what exactly is closed — on the [Licensing](/agents/proxy/license) page. ## First sign-in Installation is described in the [Installation](/agents/proxy/install) section. There is one administrator account, set with the `PROXYAGENT_ADMIN_LOGIN` and `PROXYAGENT_ADMIN_PASSWORD` environment variables. The password is stored not as-is but as its irreversible hash (SHA-256, written lowercase) — the built-in generator in the login form produces the ready string. The variables are read on every request, so the sign-in can be changed without a restart. A sign-in lasts 90 days; you will not need to re-enter the password. ## First steps 1. **Set the license key** — with the `AGENT_LICENSE` variable on the server (the key itself or a path to a file with it), then restart the container. Without a valid key the interface opens, but no traffic flows: requests to the models will be refused. 2. **Create a model** — give its public name, the vendor address, the vendor key and the prices per million “portions of text”. Without prices the product counts the model as free. 3. **Check the connection** — the “Test” button sends a request along the model’s real route and shows whether the vendor answers. 4. **Create an access key** — name it after the person or the program. The ready key is shown exactly once: save it right away. 5. **Connect one program** — in it, replace the vendor address with your server’s address, and the vendor key with your access key. 6. **Look at the log** — after the first request a row appears there: the model, the key, the cost, the duration. If there is no row, the programs are not reaching the server yet. ## What you will need Honestly, from the start: the product is not a cloud service. It needs your server with Docker and a separate PostgreSQL database — that is where the logs, spending and service data live. All of it stays with you, and only the server itself needs internet access to reach the model vendors. Next: how to issue access keys and set their limits — in the [Access Keys](/agents/proxy/docs/keys) section. [← Back to the documentation index](/agents/proxy/docs)