1 of 11

OpenRouterBridge

Setup guide for dummies

2 of 11

How to directly use openrouter without colab

With new janitorai proxy update, you can now directly use openrouter without need to setup colab proxy first.

Without colab, you lost control over some setting like minp/topp/topk, assistant prefill (in case you use claude) and auto trim feature.

Those setting are mandatory for certain model to reduce repetition or increase creativity.

but this way could be easier than dealing with colab/cloudflare issue.

  1. Select default model on openrouter
  2. Put https://openrouter.ai/api/v1/chat/completions in proxy url, and add key like this
  3. Empty the model field
  4. Save and chat

3 of 11

1.Open the colab

4 of 11

2.Click play button and wait for cloudflare link to be generated

5 of 11

3.Create your key at OpenRouter

Name: any name you like

Credit limit: maximum credit this key�can use. In case you want to give�Some money to other people.

6 of 11

4. Select model in setting page

You can change model here and it will immediately change without need to restart colab

There are several free models too (mostly 7-9b), if you don’t want to pay them yet.

llama 3.1 8b and gemma 2 9b are recommended, since both of these are newest

7 of 11

Cheap model recommendation

  • Wizard 8x22b ($0.00065 per 1k context size)

Very smart for its price. Maybe the best price per performance model in openrouter. Excel in both sfw and nsfw

  • Gemma 2 27b ($0.00027 per 1k context size)

A bit censored, but it has solid brain power.

  • Cohere: Command R+ ($0.003 per 1k context size)

Smarter than hermes. But may produce incoherent stuff sometimes, because it is not made for rp

  • Claude 3/3.5 sonnet (self-moderated) ($0.003 per 1k context size)

Very smart model from Anthropic, sonnet 3 is very good at dark theme rp like abusive or yandere. While 3.5 is very smart. (prefill_enabled must be checked on colab)

  • Fimbulvetr 11B v2 (0.0005$ per 1k context size)

Most popular rp kobold model for a while now. But it could be hosted for free with kobold colab, so I suggest using kobold more if you want to use this model

8 of 11

5. Go to your Janitor

Put your cloudflare link and your key in as openai reverse proxy. Click save and enjoy your bot

Or use https://openrouterproxy.hibikiass.uk/ as URL if you are lazy or have difficulty setting up your private proxy.�(it is on my personal server so I don’t recommend to always use it, unless you really not care about your privacy)

9 of 11

Payment

In setting tab. You can click this button,�To buy credit that used for access ai models

It will let you pay through stripe

If you want to pay with google pay, you needs�To logout from stripe first

10 of 11

Most model here are cheaper than�Openai and not as shakespeare.

It is good enough to do rp on, although�the brain is not as big as gpt4.

Bigger model(more b in its name)�usually equal to smarter model. But�writing quality is subjective and vary on�several models. Choose one you like�

*8x7b model is somewhere between 33b and 70b

11 of 11

Other guide

If you want to use 13b or smaller model. I suggest using koboldcpp colab instead, as it is free. (kobold colab can’t handle anything beyond 16b, so I only use openrouter for anything bigger than that to save some credits)

https://colab.research.google.com/drive/1l_wRGeD-LnRl3VtZHDc7epW_XW0nJvew

Guide is at the top of the page