x0root AI API - Intelligent Cloudflare Worker Playground

Chat Sessions

Chat ID: default

Model:

Context:

x0root AI API Playground

Enter your Bearer API key (`qp-...`) in the top right to start chatting with intelligent auto-routing, NVIDIA NIM waterfalls, Puter fallback, and Durable Object memory! Supports Markdown & LaTeX math rendering.

Chat: default

Shift + Enter for new line

API Explorer & Endpoint Tester

Test public and authenticated endpoints directly against `puter-api.x0root.workers.dev` using your API key.

GET /

Public health check endpoint. No API key required.

GET /v1/models

Fetch Puter model metadata in OpenAI style format. Requires Authorization Bearer key.

POST /v1/chat/completions

Primary OpenAI-compatible chat completions endpoint with intelligent auto-routing and memory.

Request JSON Body:

Intelligent Router & Tier Architecture

How x0root Worker routes prompts through classifiers, NVIDIA NIM pools, and Puter fallbacks.

High-Level Request Flow

1. Validation

API Key checked against Supabase backend; requests count incremented.

2. AI Classification

Fast NVIDIA model classifies prompt into one of 6 execution tiers.

3. Waterfall Retry

Tries NVIDIA model #1 & #2, then falls back to Puter AI provider if needed.

4. DO Memory

Saves to Cloudflare Durable Object SQLite with BM25 search and auto-compaction.

The 6 Classifier Tiers

LITE Fast / Casual

Simple questions, casual chat, rewriting, extraction, formatting, short answers.

Gemma-2-2b, Gemma-3n, Nemotron-mini

NORMAL Daily Assistant

Daily assistant tasks, normal coding help, summaries, ordinary reasoning.

Llama-3.1-8b, GPT-oss-20b, Minimax-m2.7

NORMAL_CALCULATION Math & Physics

Math, physics, engineering, finance, statistics, algorithms, standard quantitative reasoning.

Qwen3.5-122b, GPT-oss-120b, GLM-5.1

MEDIUM Nuanced Writing

Nuanced writing, multi-step analysis, medium-to-heavy coding, planning, stronger judgment.

Llama-3.3-70b, Mistral-medium, Kimi-k2.6

HEAVY_CALCULATION Advanced Proofs

Advanced proofs, graduate physics, long derivations, symbolic reasoning, difficult optimization.

Nemotron-3-ultra-550b, Mistral-large-3

HEAVY Hard Reasoning

Extremely hard non-calculation reasoning, severe debugging, large architecture, high-stakes synthesis.

Nemotron-3-ultra-550b, Llama-3.3-70b

Durable Object Memory & BM25 Search

The Worker maintains chat memory using Cloudflare Durable Objects backed by SQLite. On each user message, MiniSearch indexes older messages and retrieves relevant context using BM25. When conversation history exceeds 25 messages, automatic background compaction merges history into a structured rolling summary while preserving recent turns.

Settings & Authentication

Configure your Worker base URL and your valid Bearer API Key (`qp-...`).

Worker Base URL:

Default: https://puter-api.x0root.workers.dev

Bearer API Key (`qp-...`):

All `/v1/*` endpoints require a valid Supabase-backed API key starting with `qp-`.