Architecture

What actually runs where. Useful if you're deciding whether to depend on this API, debugging something unexpected, or just curious.

Domains

Domain / path	What it is
www.israelgpt.site	The main site: web chat and docs.
www.israelgpt.site/api/v1/chat/completions, /models	The OpenAI-compatible API (the base URL is .../api/v1; .../v1 works too).
www.israelgpt.site/api/v1/chat	The native API, with personas.
www.israelgpt.site/dashboard	Your API keys and their request logs.
www.israelgpt.site/docs	This documentation.

Always use www.israelgpt.site. No other hostname is supported.

Request flow

your server / bot

│ HTTPS

▼

www.israelgpt.site (edge network)

│ pages are static files; API calls go on, over a private link

▼

API worker (no public address of its own)

├──▶ the key, rate limits and safety checks

├──▶ database: keys (hashed), your request log, encrypted chat logs

└──▶ the model: the actual completion

▼

the reply, back to you

Services involved

Service	Role
Hosting and security provider	Hosts everything: the site, the API worker, rate limiting and the database. There is no other hosting, database or cache provider.
Model hosting	IsraelBot 1 is our own model, trained and tuned in-house (see Personas & Models). YahooBot 1 is a fine-tuned DeepSeek V4, served through OpenRouter. Only the API worker talks to either.
Discord	Only relevant if you use or fork the reference Discord bot (see Discord Bot). It runs as a separate process outside this stack.

What this means for you

Each request stands alone. There is no server-side memory between calls; send the conversation you want the model to see (on the native endpoint, with each reply's sig).
Web pages can call it (CORS is on), but a key in a page is a vulnerability: anyone can copy it. Where you can, call it from a server.
Model generation is almost all of the time a request takes.
If the model fails to respond, the request fails with 502. There is no switch to a different model. See Errors.
None of this infrastructure is exposed to you directly. You only ever talk to www.israelgpt.site.

Streaming

The OpenAI-compatible endpoint streams: yahoobot-1's reply is passed on as the model writes it, in OpenAI's event format, with small batches about ten times a second to stay inside the hosting plan's CPU budget. israelbot-1's reply is finished first so it can be checked as a whole, then sent as a stream.

The native endpoint doesn't stream, on purpose: the reference client (a Discord bot) posts one finished message rather than editing it token-by-token, and a plain JSON contract keeps onboarding to a curl command and a 10-line snippet.

// native /chat: one response, once generation finishes
const data = await res.json();
console.log(data.reply);