Examples

The basic request, plus a few common patterns: multi-turn context, picking a persona, handling media, and retrying safely.

Basic request

curl -X POST https://www.israelgpt.site/api/v1/chat \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{ "role": "user", "content": "What time is it in Tel Aviv?" }],
    "personaId": 1,
    "selectedModel": "israelbot-2",
    "effort": "medium"
  }'

"Give to your AI agent" copies a full plain-text spec of this API (auth, parameters, response shape, errors) — paste it into Claude Code, Cursor, or any coding assistant and ask it to wire up the integration for you.

Multi-turn conversation

There's no server-side conversation memory for the API — you send the full history you want considered on every call, oldest message first. The reference Discord bot does this by reading the real last 10 Discord messages in a channel on every trigger (see Discord Bot).

{
  "messages": [
    { "role": "user", "content": "What's the capital of Israel?" },
    { "role": "assistant", "content": "Jerusalem, obviously..." },
    { "role": "user", "content": "And the largest city?" }
  ]
}

Picking a persona and model

See the full lists on Personas & Models. Example: the Tsundere persona (id 9) on IsraelBot 1.5, high effort:

{
  "messages": [{ "role": "user", "content": "Help me plan a trip to Eilat." }],
  "personaId": 9,
  "selectedModel": "israelbot-1.5",
  "effort": "high"
}

Uncensored mode

{
  "messages": [{ "role": "user", "content": "Roast my code review." }],
  "uncensoredMode": true
}

The Terms of Service's content and eligibility rules apply regardless of this flag — it changes language, not what's allowed.

Handling media in the response

If the persona's reply used an [IMAGE], [car], [podcast], or [music] tag, media carries ready-to-use URLs — no need to parse the raw tags yourself:

const data = await res.json();

if (data.media?.images) {
  for (const img of data.media.images) {
    console.log("image:", img.url, img.tags);
  }
}
if (data.media?.podcast) {
  console.log("podcast (audio bytes at this URL):", data.media.podcast.url);
}

Retrying safely

A minimal retry loop that backs off on 429/502/503 and gives up on anything else (a 400/401 won't fix itself by retrying):

import httpx
import time

def ask(messages, max_attempts=3):
    for attempt in range(max_attempts):
        response = httpx.post(
            "https://www.israelgpt.site/api/v1/chat",
            headers={"Authorization": "Bearer YOUR_API_KEY"},
            json={"messages": messages},
            timeout=60,
        )
        if response.status_code == 200:
            return response.json()

        if response.status_code == 429:
            wait = response.json()["error"].get("retry_after_seconds", 5)
        elif response.status_code in (502, 503):
            wait = 2 ** attempt  # simple exponential backoff
        else:
            response.raise_for_status()  # 400/401 - won't fix itself, stop

        time.sleep(wait)

    raise RuntimeError("Gave up after retries")

Stripping bracket tags for a plain-text UI

If you're building something that only wants clean prose (no [ACTION:...]/[MEMORY:...]/etc. left visible), a simple regex strip works — this is exactly what the reference bot does:

import re

STRIP_TAG_RE = re.compile(
    r"\[(ACTION:[^\]]*|MEMORY:[^\]]*|flashcard:[^\]]*|QUESTION:[^\]]*|"
    r"get-time:[^\]]*|get-date:[^\]]*|create-file:[^\]]*|"
    r"LGBTQ_IMG|SEND_IDF|download|crisis-event[^\]]*)\]",
    re.IGNORECASE,
)

clean_text = STRIP_TAG_RE.sub("", data["reply"]).strip()