Client setup

Menus and labels move around between versions of these tools, so match the names above loosely. If a tool lets you type any base URL and any model name, it will work. If it insists on a list of known models, use its "OpenAI Compatible" option.

The free plan is for chat, not for coding agents. An agent such as Cline or Roo Code sends a system prompt of 10,000 tokens or more before it says anything, which doesn’t fit in 12,000 tokens, and it makes many requests for one task. The API will tell you so (context_length_exceeded, see Sizes, safety and errors) and the tool will complain about the context. For agents, apply for the Pro API.

1. Open the tool’s settings and pick the provider OpenAI Compatible.
2. Base URL https://www.israelgpt.site/api/v1, your key, Model ID yahoobot-1.
3. Open the model configuration in the same panel. Set the context window and max output tokens to your plan’s numbers, and turn image support off. With the right context window the tool shortens its own history before it gets to the limit.
4. Leave native tool calling alone unless you know you want it. By default these tools put their tools in the prompt as XML, which needs nothing special from the API. If you do turn on native tool calling, it works with yahoobot-1 (see Tool calling).

Their requests are long and arrive often. Your plan’s requests a minute are what limits an agent: Cline stops and waits when it gets a 429. Your daily tokens are the other limit; the dashboard shows both.

In its config file, a model with the OpenAI provider and our base URL:

models:
  - name: IsraelGPT YahooBot
    provider: openai
    model: yahoobot-1
    apiBase: https://www.israelgpt.site/api/v1
    apiKey: igpt_...
    defaultCompletionOptions:
      contextLength: 64000
      maxTokens: 8000

Older versions use a JSON config with "provider": "openai", "apiBase", "apiKey", "contextLength" and "maxTokens" under the model. Use your plan’s numbers.

export OPENAI_API_BASE=https://www.israelgpt.site/api/v1
export OPENAI_API_KEY=igpt_...
aider --model openai/yahoobot-1

The openai/ prefix tells Aider (through LiteLLM) to send the rest of the name to an OpenAI-compatible server. Aider doesn’t know this model, so it may warn about its context size; it still works. Its prompts are large, so use a Pro plan.

In the admin settings, under Connections, add an OpenAI API connection: URL https://www.israelgpt.site/api/v1, your key. The models list (GET /models) fills in yahoobot-1 and israelbot-1 on its own, and costs nothing against your limits. The same goes for LibreChat, SillyTavern, Chatbox and similar apps: a custom OpenAI endpoint with the URL and key above.

For long chats, choose yahoobot-1: israelbot-1 sees earlier replies as quotes, since an OpenAI client can’t send back IsraelGPT’s signatures.

# Python:  pip install openai
from openai import OpenAI
client = OpenAI(api_key="igpt_...", base_url="https://www.israelgpt.site/api/v1")

# JavaScript:  npm install openai
import OpenAI from "openai";
const client = new OpenAI({ apiKey: "igpt_...", baseURL: "https://www.israelgpt.site/api/v1" });

Anything built on these libraries (LangChain, LlamaIndex, Vercel AI SDK's OpenAI provider and so on) takes the same two settings. Keep the key on a server.

More: the error list and the key’s request log on the dashboard, which shows every request and its result.