OpenAI-compatible · your keys · encrypted memory

Give any API model a memory.

Point your OpenAI-compatible client at one endpoint. ModelRouters adds a persistent, editable memory layer — your model remembers across sessions, stays under context limits, and costs less. Bring your own key, or use a hosted model.

30 days free with no card · then optional launch sale: $6/mo $3/mo for memory · bring-your-own-key routing is free · hosted models pay-as-you-go

Endpoint
Base URL
https://api.modelrouters.net/v1
Chat completions
https://api.modelrouters.net/v1/chat/completions
Drop-in

Works with the OpenAI SDK, LangChain, or plain curl — just change the base URL and use your ModelRouters key.

How it works

1

Point to one endpoint

Set your OpenAI-compatible client's base URL to https://api.modelrouters.net/v1. No SDK swap, no rewrite.

2

Bring your key, or use hosted

Connect DeepSeek or another direct provider, name your models, and route them for free — or use a hosted model with credits.

3

Memory does the rest

Facts are distilled, injected, and compressed as the chat grows. It's yours — view, edit, or delete any of it, any time.

Why a memory layer

The model stays the same. What changes is that it remembers.

Persistent memory

Your model remembers across sessions — not just within a single context window.

Context compression

Engages before you hit the limit, keeping long chats coherent and cutting token cost.

Yours to edit

User-visible, editable memory and chat history. No black box — read it, change it, delete it.

Encrypted

Per-account encryption at rest. Automated-only handling — no human review of your content.

Any model

Works with any OpenAI-compatible endpoint. Switch providers without losing your memory.

One key, one endpoint

A single ModelRouters key routes to every provider you connect. Drop-in and streaming.

Supported models

Use every provider hosted or bring your own key. Every model gets the same memory layer over the same OpenAI-compatible API.

OpenAI

Hosted

The complete GPT-5.6 family: Luna, Terra, and Sol.

gpt-5.6-lunagpt-5.6-terragpt-5.6-sol
Hosted + BYOK

Claude

Hosted

Haiku, Sonnet, Opus, and Fable — including the Claude 5 family.

claude-haiku-4.5claude-sonnet-4.6claude-sonnet-5claude-opus-4.8claude-opus-5claude-fable-5
Hosted + BYOK

Grok

Hosted

Six text models from Grok 4.20 through Grok 4.6 and Grok Build.

grok-build-0.1grok-4.3grok-4.20-0309-non-reasoninggrok-4.20-0309-reasoninggrok-4.5grok-4.6
Hosted + BYOK

DeepSeek

Hosted

V4 Flash with default, fast, and reasoning modes, plus V4 Pro.

deepseek-v4-flashdeepseek-v4-flash-fastdeepseek-v4-flash-reasoningdeepseek-v4-pro
Hosted + BYOK

Z.AI

Hosted

Eight GLM models from 4.5 Air through the latest GLM-5.2.

glm-4.5-airglm-4.5glm-4.6glm-4.7glm-5glm-5-turboglm-5.1glm-5.2
Hosted + BYOK

Pricing

One flat price for memory and saved chats. You only pay for inference you actually run.

Memory + saved chats
Launch sale
$6$3 / mo

First 30 days free with no card. The trial ends automatically; subscribe afterward for $3/month to continue.

  • • Persistent, editable memory on any model
  • • Context compression + encrypted chat history
  • • One cardless free trial per account
Start 30-day trial
Bring your own key
Free routing
  • • DeepSeek recommended; other provider connections supported
  • • You pay your provider directly
  • • Encrypted at rest; decrypted only to call your provider
Hosted inference
Credits / as you go
  • 27 current models across five providers
  • • At provider cost + a thin buffer
  • • Top up any amount