Quickstart

Knox is a unified, OpenAI-compatible API, Anthropic Messages API format supporting and platform that gives you access to hundreds of models through one endpoint, with intelligent routing, fallbacks, and transparent pricing. It pairs that with Knox‑MS, a memory‑centric system for long‑running agents and apps that need persistent context.

What you get

Smart model routing: performance/cost/balanced strategies with automatic fallback and provider metrics.
Multimodal I/O: text, images, PDFs, and audio inputs, plus image, video, and music generation, text to speech, speech to text, voices, voice changer, Voice Isolator, sound effects, dubbing, forced alignment, and speech-to-speech Realtime.
Tools and MCP: OpenAI‑compatible tool calling and MCP server bridging.
Web search: :online model variant or the web plugin for live citations.
Jev (System One): typed questions against a shared state, structured answers (not an LLM). See Jev.
Structured & reasoning outputs: JSON Schema enforcement and standardized reasoning tokens.
Efficiency features: prompt caching and middle‑out message transforms for long contexts.
Reliability: zero‑completion insurance so failed/empty responses aren’t billed.

Knox‑MS Memory System

Knox‑MS adds multi‑level memory, summaries, vector search, and a knowledge graph to power persistent agents and workflows. It also supports autonomous planning, task orchestration, self‑healing, and realtime progress events. Learn more in the Knox Memory System — Full Feature Deep Dive.

Using the OpenAI SDK

TypeScript
Python

import OpenAI from 'openai';

const openai = new OpenAI({
  baseURL: 'https://api.knox.chat/v1',
  apiKey: '<KNOXCHAT_API_KEY>',
});

async function main() {
  const completion = await openai.chat.completions.create({
    model: 'knox/knox-ms',
    messages: [
      {
        role: 'user',
        content: 'What is the meaning of life?',
      },
    ],
  });

  console.log(completion.choices[0].message);
}

main();

from openai import OpenAI

client = OpenAI(
  base_url="https://api.knox.chat/v1",
  api_key="<KNOXCHAT_API_KEY>",
)

completion = client.chat.completions.create(
  model="knox/knox-ms",
  messages=[
    {
      "role": "user",
      "content": "What is the meaning of life?"
    }
  ]
)

print(completion.choices[0].message.content)

Using the Knox.Chat API directly

Python
TypeScript
Shell

import requests
import json

response = requests.post(
  url="https://api.knox.chat/v1/chat/completions",
  headers={
    "Authorization": "Bearer <KNOXCHAT_API_KEY>",
  },
  data=json.dumps({
    "model": "anthropic/claude-sonnet-4.6", # Optional
    "messages": [
      {
        "role": "user",
        "content": "What is the meaning of life?"
      }
    ]
  })
)

fetch('https://api.knox.chat/v1/chat/completions', {
  method: 'POST',
  headers: {
    Authorization: 'Bearer <KNOXCHAT_API_KEY>',
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    model: 'anthropic/claude-sonnet-4.6',
    messages: [
      {
        role: 'user',
        content: 'What is the meaning of life?',
      },
    ],
  }),
});

curl https://api.knox.chat/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $KNOXCHAT_API_KEY" \
  -d '{
  "model": "anthropic/claude-sonnet-4.6",
  "messages": [
    {
      "role": "user",
      "content": "What is the meaning of life?"
    }
  ]
}'

The API also supports streaming.

Using third-party SDKs

For information about using third-party SDKs and frameworks with Knox, please see our frameworks documentation.

Principles

Knox helps teams build reliable, cost‑efficient SI systems across providers. We believe the future is multimodal, multi‑provider, and memory‑centric.

Why Knox?

Price and Performance. Knox scouts for the best prices, lowest latencies, and highest throughput across providers, and lets you choose how to prioritize them.

Standardized API. No code changes when switching between models or providers. Use OpenAI‑compatible SDKs, tools, and structured outputs out of the box.

Memory‑First Apps. Knox‑MS enables long‑running agents with persistent memory, knowledge extraction, and autonomous planning.

Multimodal by Default. Images, PDFs, and audio inputs plus image generation, video generation, and speech-to-speech Realtime make multimodal workflows straightforward.

Consolidated Billing. Simple and transparent billing, regardless of how many providers you use, with zero‑completion insurance for failed/empty generations.

Higher Availability. Automatic fallback and smart routing keep requests working even when providers go down.

Models

One API for hundreds of models

Explore and browse 300+ models and providers on our website, or with our API.

Models API Standard

Our Models API makes the most important information about all LLMs freely available as soon as we confirm it.

API Response Schema

The Models API returns a standardized JSON response format that provides comprehensive metadata for each available model. This schema is cached at the edge and designed for reliable integration for production applications.

Root Response Object

{
  "data": [
    /* Array of Model objects */
  ]
}

Model Object Schema

Each model in the data array contains the following standardized fields:

Field	Type	Description
`id`	`string`	Unique model identifier used in API requests (e.g., `"google/gemini-2.5-pro"`)
`object`	`string`	Object type identifier (always `"model"`)
`created`	`number`	Unix timestamp of when the model was created
`owned_by`	`string`	Organization that owns the model
`permission`	`ModelPermission[]`	Array of permission objects defining access controls
`root`	`string`	Root model identifier
`parent`	`string \| null`	Parent model identifier if this is a fine-tuned version
`context_length`	`number`	Maximum context window size in tokens
`architecture`	`Architecture`	Object describing the model's technical capabilities
`pricing`	`Pricing`	Price structure for using this model
`top_provider`	`TopProvider`	Configuration details for the primary provider
`supported_parameters`	`string[]`	Array of supported API parameters for this model

Architecture Object

{
  "modality": string, // High-level description of input/output flow (e.g., "text+image->text")
  "input_modalities": string[], // Supported input types: ["file", "image", "text", "audio"]
  "output_modalities": string[], // Supported output types: ["text"]
  "tokenizer": string // Tokenization method used (e.g., "Gemini")
}

Pricing Object

All pricing values are in USD per token/request/unit. A value of "0" indicates the feature is free.

{
  "prompt": string, // Cost per input token
  "completion": string, // Cost per output token
  "request": string, // Fixed cost per API request
  "image": string, // Cost per image input
  "audio": string, // Cost per audio input
  "web_search": string, // Cost per web search operation
  "internal_reasoning": string, // Cost for internal reasoning tokens
  "input_cache_read": string, // Cost per cached input token read
  "input_cache_write": string // Cost per cached input token write
}

Top Provider Object

{
  "context_length": number, // Provider-specific context limit
  "max_completion_tokens": number // Maximum tokens in response
}

Supported Parameters

The supported_parameters array indicates which OpenAI-compatible parameters work with each model:

include_reasoning - Include reasoning in response
max_tokens - Response length limiting
reasoning - Internal reasoning mode
response_format - Output format specification
seed - Deterministic outputs
stop - Custom stop sequences
structured_outputs - JSON schema enforcement
temperature - Randomness control
tool_choice - Tool selection control
tools - Function calling capabilities
top_p - Nucleus sampling

:::note Token Counting Differences Different models use different tokenization methods (as indicated by the tokenizer field in the model schema). Some models break up text into chunks of multiple characters (GPT, Claude, Llama, etc), while others tokenize differently (like Gemini). This means that token counts (and therefore costs) will vary between models, even when inputs and outputs are the same. Costs are displayed and billed according to the tokenizer for the model in use. You can use the usage field in API responses to get the actual token counts for your input and output. :::

Quickstart

What you get

Knox‑MS Memory System

Using the OpenAI SDK

Using the Knox.Chat API directly

Using third-party SDKs

Principles

Why Knox?

Models

Models API Standard

API Response Schema

Root Response Object

Model Object Schema

Architecture Object

Pricing Object

Top Provider Object

Supported Parameters

Frequently Asked Questions

Getting started

Models and Providers

API Technical Specifications

Privacy and Data Logging

What you get​

Knox‑MS Memory System​

Using the OpenAI SDK​

Using the Knox.Chat API directly​

Using third-party SDKs​

Principles​

Why Knox?​

Models​

Models API Standard​

API Response Schema​

Root Response Object​

Model Object Schema​

Architecture Object​

Pricing Object​

Top Provider Object​

Supported Parameters​

Frequently Asked Questions​

Getting started​

Why should I use Knox?

How do I get started with Knox?

What is Knox‑MS and when should I use it?

What is Jev and when should I use it?

Where can I get technical support?

How do I pay for using Knox?

How do I recharge?

Models and Providers​

What LLM models does Knox support?

How frequently are new models added?

What are model variants?

What is the expected latency/response time for different models?

How does model fallback work if a provider is unavailable?

API Technical Specifications​

What authentication methods are supported?

How are rate limits calculated?

What API endpoints are available?

What are the supported formats?

How does streaming work?

What SDK support is available?

Privacy and Data Logging​

What data is logged during API use?

What third-party sharing occurs?

What data is logged during Chat page use?

What you get

Knox‑MS Memory System

Using the OpenAI SDK

Using the Knox.Chat API directly

Using third-party SDKs

Principles

Why Knox?

Models

Models API Standard

API Response Schema

Root Response Object

Model Object Schema

Architecture Object

Pricing Object

Top Provider Object

Supported Parameters

Frequently Asked Questions

Getting started

Models and Providers

API Technical Specifications

Privacy and Data Logging