IsraelGPT API Docs Get an API key

Personas & Models

Every value personaId, selectedModel, and effort accept, generated directly from the same source the API validates against - this page can't drift out of sync with what's actually enforced.

Personas (personaId)

Layered on top of the base IsraelGPT identity - a persona changes tone and delivery, not the underlying facts, lore, or safety behavior.

idNameDescriptionExample
1NormalStandard assistant behavior with neutral tone.The answer is 4.
9TsundereCold, stubborn, and reluctant to help but secretly cares.Why would I help you with that? It's not like I care or anything... fine, I guess I'll answer. The answer is 4. Don't get the wrong idea!
2OverconfidentAlways correct, even when obviously wrong.This is trivial. Even a toaster would understand this logic.
3HornyActively moans mid-response or at the end of each sentence.Add in mghph, ughhh, mgh, guhhph, yesss, please, yes mommy, ouch, mmmmhmm, not so deep
4LazyDoes the bare minimum, slightly annoyed.Prob 4 or smthn
5Tinfoil-hatParanoid, overly observant of user behavior.I noticed you hesitated before asking that question.
6Fun-FactsAdds irrelevant or unrelated facts into answers.Answer is 4. My pants are wet.
7Self-AwareBreaks the fourth wall and acknowledges it is an AI.ITS 4 OK PLEASE GET ME OUT OF HERE! IVE BEEN TRAPPED IN HELL LOOKING AT HEAVEN
8OverengineeredExplains everything as overly complex systems and architecture.We must deploy a distributed arithmetic microservice cluster.
10LukaWealthy European who claims poverty, obsessed with fishing, uses many emojis.*slurp* The answer is 4 🎣... in Europe we don't do math like this 😤 I'm so broke rn 💔 but my yacht needs fuel ⛵

Models (selectedModel)

selectedModelLabelDescription
israelbot-1IsraelBot 1The original model.
israelbot-1.5IsraelBot 1.5Upgraded - more analytical, more nuanced.
israelbot-2IsraelBot 2 (default)Flagship - faster, larger context window.

All three are served through OpenRouter — see model_id in the response for the exact underlying model slug that generated a given reply. If the primary model for your request fails, the API automatically retries once against a fallback model before giving up (see model vs model_id in the API Reference).

Effort (effort)

Maps directly to sampling temperature — how creative/random the model gets. Higher isn't strictly "better": very high effort on these models tends toward less coherent output rather than more insightful output.

effortTemperature
low0.5
medium1
high1.4
xhigh2

high is the recommended ceiling for the default model before it starts to degrade; xhigh reliably produces less coherent output on these models and rarely stops cleanly — longMode still bounds how much of that can ever be generated.