how to config AI for selfhost

[IamTaoChen]

Can we use AI in the self-host instance?

[wanghkkk]

I have the same question

[ngxnvux]

Turn on the copilot plugin at
.affine/self-host/config/affine.js

[lennon101]

Turn on the copilot plugin at
.affine/self-host/config/affine.js

nice find and thanks for this! I've gone ahead and uncommented the copilot plugin and added my OpenAI apikey (see below). And I in the Affine logs I can see the CoPilot is registered. But I'm still getting an error. Do I need to get a FAL and an unsplashKey as well?

affine.js

// /* Copilot Plugin */
AFFiNE.use('copilot', {
  openai: {
    apiKey: 'sk-proj-<withheld>',
  },
  // fal: {
  //   apiKey: 'your-key',
  // },
  // unsplashKey: 'your-key',
  // storage: {
  //   provider: 'cloudflare-r2',
  //   bucket: 'copilot',
  // }
})

Affine logs:

[Nest] 117  - 06/26/2024, 2:23:10 AM VERBOSE [CacheInterceptor] cache blobs:6ebf69e5-33b9-4e3b-9809-a99de7acab75 miss
[Nest] 117  - 06/26/2024, 2:23:18 AM   DEBUG [CopilotProviderService] Copilot provider openai registered, capabilities: text-to-text, text-to-embedding, text-to-image, image-to-text

UI error when trying to use AI:

[Bottleface]

@lennon101 were you ever able to get this working?

[AndreasWidmer]

For everybody interested: this works with the self hosted version. You have to use (a paid) Open AI account with a User API keys (Legacy) - not the new project API keys!

In the affine.js config set:

AFFiNE.use('copilot', {
   openai: {
           apiKey: 'your-key',
   },
//   fal: {
//     apiKey: 'your-key',
//   },
//   unsplashKey: 'your-key',
//   storage: {
//     provider: 'cloudflare-r2',
//     bucket: 'copilot',
//   }
 })

[campbellmcgregor]

any idea how to remove the 10 usage limit?

[madratzz]

You can modify it in the postgres database

[pyrox82]

Can I also choose which model is used? I think by default it's gpt-4o. But I rather want to use gpt-4o-mini.

[JoeBuildsStuff]

You might need to modify the model selection in the postgres table

[richardy2012]

At this stage, modifications from the database are no longer effective, and restrictions are made directly at the code level. Only a series of built-in large models such as gpt-4o and gpt-4o-mini are supported, so don't bother with it.

[Nishan-Vivek]

Looks like openai don't allow new accounts to generate legacy API keys anymore. Anyway to get this to work with the project api keys?