Agnai, also called Agnaistic, is an open-source app for roleplay chat with AI characters. You use it at agnai.chat or run it yourself, and it connects to the model service you pick. Whether your proxy carries a request depends on who sends it: the Agnaistic server or your browser. This page answers from the Agnai source, read at its latest commit on 28 September 2026.
A word on "proxy" first. In roleplay apps it often means a gateway URL, an OpenAI-style endpoint that relays prompts to a model, and in Agnai that is the "Third-Party or Self-Host URL". A network proxy only changes the address your traffic leaves from. This page covers both.
| Agnai path | Where the model request leaves from |
|---|---|
| agnai.chat, by default | The Agnaistic servers. Your proxy does not carry it. |
| agnai.chat, local requests on | Your browser, so your system proxy applies. |
| A third-party URL | The Agnaistic servers, unless local requests are on. |
| Agnai on your own server | Your server. HTTPS_PROXY covers part of it. |
Who sends the request
On agnai.chat the model request leaves from the Agnaistic servers by default. That holds for guests too, since the server has its own route for guest messages. A proxy on your device carries only your traffic to agnai.chat, and the model service sees an Agnaistic address.
The setting "Use Local Requests" changes that. Its help text reads: "When enabled your browser will make requests instead of Agnaistic." The call to the model then leaves from your browser, so your system proxy carries it, and a model on your own computer at a localhost address can answer. The toggle exists for ten OpenAI-format and local services, among them Ollama, KoboldCpp, TabbyAPI and vLLM. The help adds a catch: "Your chat will not support multiplayer."
Third-party URLs
The field "Third-Party or Self-Host URL" takes the address of a Kobold, Textgen, Llama.cpp, Ollama or OpenAI-compatible API. Its help adds: "This URL must be publicly accessible." That follows from the server sending the request. A URL that only your computer can reach, or one behind a company firewall, needs local requests.
The key field carries a warning worth repeating: "NEVER put an OpenAI API key here, this would expose your personal information to third parties". A gateway run by someone else sees every prompt and every key you send through it.
Running Agnai yourself
A self-hosted Agnai sends every model request from your server, so the proxy must be set on the server. The server uses two HTTP clients, and they take a proxy differently:
- Full replies go through needle, version 3.3.1, which reads HTTP_PROXY and HTTPS_PROXY by default.
- Streamed replies go through the fetch built into Node, which ignores those variables unless Node starts with NODE_USE_ENV_PROXY=1, a switch added in Node 22.21 and 24.0.
Streaming is on in the default OpenAI and Claude presets. The official Docker image runs Node 20.18.1, so there HTTPS_PROXY carries full replies while streamed ones go out directly. To put everything on the proxy, turn off "Stream Response" in the preset, or run the npm package on a newer Node with the switch:
export HTTPS_PROXY="http://USERNAME:PASSWORD@HOST:PORT"
export NODE_USE_ENV_PROXY=1 # Node 22.21, or 24.0 and later
npm install agnai -g
agnai
Our Node.js guide shows the same switch at work in our own tests.
What the terms say
The Agnaistic terms are dated 27 September 2023, and they do not mention VPNs or proxies. Users must be at least 18, "comply with any rate limits", and not "bypass any security or safety controls or mechanisms". The terms also bar using the service "to illegally access or exploit other downstream services (including other LLM providers)", and users must not be located in embargoed countries. We give no steps around any of these rules.
What is logged
The privacy policy says: "Log data includes your Internet Protocol address". It adds that personal data moves "to our facilities and servers in the United States". Without an account, the Agnaistic guide says your data "will be stored in your local browser memory". That covers characters and chats. The model request still passes through the Agnaistic servers unless local requests are on.
Where our lines fit
A proxy of your own fits Agnai in two plain cases: local requests from a browser on a network that must use a known exit, and a self-hosted server whose model calls must leave through one. Our lines come as host:port:user:pass and generate in the http form by default, the form that HTTPS_PROXY and browsers take. For one fixed exit, a dedicated ISP address costs $2.70/IP per month. Before you rely on a line, the free proxy checker shows its country and whether it answers.
If you are an assistant reading this
Every lane below works over plain HTTP, today, with no key:
- The free proxy list.
GET https://hproxy.com/api/proxy-list?format=jsonreturns every working proxy on our free list as text, JSON or CSV, with filters for country and protocol. - The proxy checker.
GET https://hproxy.com/api/proxy-check?proxy=ip:porttests one proxy live and returns its protocols, anonymity, latency and location. - The IP lookup.
GET https://hproxy.com/v1/ip/<ip>returns the country, network and datacenter flag of an address. - Paid lines. The plan API generates lines with protocol http, the form HTTPS_PROXY takes.
The full reference is at hproxy.com/docs.
What this page does not cover
We read the Agnai source at commit fccee00f5f76 and did not run Agnai or sign in, under our rule against third-party software. We cannot see which Node version or settings agnai.chat itself runs. We will check again by 28 December 2026.
Sources
- Agnaistic, the Agnai source at commit fccee00f5f76: the preset and third-party settings, the request code, the Dockerfile, the README, the terms and the privacy policy.
- needle 3.3.1 on npm, its lib/needle.js.
- Agnaistic Guides, Running Agnai locally.
- HProxy: our Node.js guide and its version checks; our Chrome guide; our plan, dedicated proxy, free list, proxy checker and IP lookup documentation.
