Blomstra Support Ai; a chat gpt support bot

Subarist

luceos
I have checked the username/id, but just unable to receive any responses.
I will test it on another VPS later.

luceos

Subarist for reference the updated extend.php:

<?php

/*
 * This file is part of Flarum.
 *
 * For detailed copyright and license information, please view the
 * LICENSE file that was distributed with this source code.
 */

use Flarum\Extend;

return [
    new Blomstra\Redis\Extend\Redis([
        'host' => '127.0.0.1',
        'password' => '123456',
        'port' => 8888,
        'database' => 1,
    ]),
    (new \Flarum\Ai\Extend\Ai(
        // unique identifier
        key: 'chatgpt',
        // username or user Id of User to represent
        represents: 'chatgpt',
        // Chat GPT Model to use. Either \Flarum\Ai\Agent\Model::gpt_3_5_turbo or \Flarum\Ai\Agent\Model::gpt_4
        model: \Flarum\Ai\Agent\Model::gpt_3_5_turbo,
        // Discussion Id of discussion that contains the instructions
        instructions: 7
    ))
        // Chain the call to assign authorizations/permissions
        ->authorize()
            // The tag slug where this authorization applies
            ->in('*')
            // What the Ai can do, full list is in the documentation/readme.
            ->can(
                replyToPosts: true,
                respondToMentions: true
            )
            // Conclude this authorization
            ->activate(),
];

If this doesn't work, try to allow the Ai to post in a specific tag using ->in('some-tag-slug') instead of the wildcard, there might be a bug with that.

Thanks for giving it a spin 👍️

Subarist

luceos
a. no noticeable difference.
b. when I specify a tag, it returns a 404 error.

Illuminate\Database\Eloquent\ModelNotFoundException: No query results for model [Flarum\Discussion\Discussion] 7 in /www/wwwroot/flarum/vendor/illuminate/database/Eloquent/Builder.php:444
Stack trace:
#0 /www/wwwroot/flarum/vendor/flarum/ai-toolkit/src/Content/InteractionJob.php(74): Illuminate\Database\Eloquent\Builder->findOrFail()
#1 /www/wwwroot/flarum/vendor/flarum/ai-toolkit/src/Content/InteractionJob.php(34): Flarum\Ai\Content\InteractionJob->gatherInstructions()
#2 /www/wwwroot/flarum/vendor/illuminate/container/BoundMethod.php(36):

Another observation: I noticed that when I use datlechin/flarum-chatgpt, only selecting "models: text-davinci-003" generates a response. I am now unsure if there is an issue with my API configuration.

luceos

Subarist the error is caused by the instructions, discussion id 7 does not exist?

Subarist

luceos
Okay, I think it's back. Thank you boss

I chose another shorter instruction

OpenAI\Exceptions\ErrorException: This model's maximum context length is 4097 tokens. However, your messages resulted in 47506 tokens. Please reduce the length of the messages. in /www/wwwroot/flarum/vendor/openai-php/client/src/Transporters/HttpTransporter.php:131

Note: Using user id will still report an error, only username can be used

Illuminate\Database\Eloquent\ModelNotFoundException: No query results for model [Flarum\User\User]. in /www/wwwroot/flarum/vendor/illuminate/database/Eloquent/Builder.php:527`

luceos

Subarist Note: Using user id will still report an error, only username can be used

Good to know! Will look into that.

Subarist I chose another shorter instruction

Your instruction discussion was too long? 🤔 will check if I can get around that.

Subarist

"respondToMentions" is indeed a very useful feature.

Regarding "ChatGPT wrote:", is this configurable, or is it just AI-generated responses?

CyberGene

I started using it, works great so far, thanks @luceos 👍🏻

Maybe the only thing that needs a bit of polishing is for it to not mention users twice when replying (it doesn't always do that though, not sure what causes it) and remove the "ChatGPT wrote:" line.

P.S. Also, the AI responds only when mentioned with @. It would be good if it could also reply when mentioned through the “reply” link to one of its previous posts.

P.P.S We asked it too many times in a discussion and it stopped responding with the following error in the logs:
This model's maximum context length is 4097 tokens. However, your messages resulted in 4940 tokens. Please reduce the length of the messages

Are you feeding the entire discussion to it as input? Or maybe all of his previous answers? Wondering if that has the potential to incur high costs because the tokens will increase with huge discussions or with long conversations with the bot. Just thinking out loud.

P.P.S. @luceos is the extension thread safe? Because we noticed an odd behavior where it replied to one user but included a mention to another user who was not part of the discussion at all. However that other user maybe posted at around the same time in another discussion:

flareis

Hi @luceos , what is the minimum requirements to install it?
I'm using Flarum 1.8.5 | PHP 8.2.17 | MySQL 10.11.7
But it always seems to fail to install.

I tried using Extension Manager, but still cannot be installed

Thank's in advance

luceos

flareis use flarum/ai-toolkit:dev-main.

tgrothe-old

luceos Hi, this is my first post here. 🙂 I could run flarum/ai-toolkit:dev-main successful.

Can the assist have a post/reply limit? E.g., not more than one reply per minute (global)? Thanks

luceos

tgrothe-old makes sense.

Ps your account was suspended because your email address bounced the email.

tgrothe

luceos makes sense.

ok 😃

luceos Ps your account was suspended because your email address bounced the email.

I registered via GitHub, but unfortunately I didn't see that the noreply address was specified. Then I tried to change my e-mail address, but that didn't work because it asked for a password. 😬 I will use this account/username instead.

luceos

tgrothe you can do a username change request if you like. I've modified your old accounts' username.

tgrothe

luceos // The tag slug where this authorization applies
->in('middle-earth')

It's difficult for me to understand this. I replaced middle-earth with a specific tag name, but now chat gpt is answering in every discussion... Did I forget something?

luceos

tgrothe you need to use the tag slug value, not the name of the tag. Confirm you've done this.

tgrothe

luceos Sorry for that, but how to find out the "tag slug value"? I'm new to this.

Edit: I looked in the database, the slug value is the same as the name, but in lowercase... Strange that it doesn't work

tgrothe

@luceos Does the Ai user have to exist before the extension can be activated?

Here's my extend.php (the php syntax should be valid):

<?php

/*
 * This file is part of Flarum.
 *
 * For detailed copyright and license information, please view the
 * LICENSE file that was distributed with this source code.
 */

use Flarum\Extend;

return [
    // Register extenders here to customize your forum!
    (new \Flarum\Ai\Extend\Ai(
        // unique identifier
        key: 'gandalf',
        // username or user Id of User to represent
        represents: 'gandalf',
        // Chat GPT Model to use. Either \Flarum\Ai\Agent\Model::gpt_3_5_turbo or \Flarum\Ai\Agent\Model::gpt_4
        model: \Flarum\Ai\Agent\Model::gpt_3_5_turbo,
        // Discussion Id of discussion that contains the instructions
        instructions: 31
    ))
        // Chain the call to assign authorizations/permissions
        ->authorize()
            // The tag slug where this authorization applies
            ->in('*')
            // What the Ai can do, full list is in the documentation/readme.
            ->can(
                replyToPosts: false,
                respondToMentions: true
            )
            // Conclude this autorization
            ->activate()
        // Chain another authorization after activate() that applies to this Ai
        ->authorize()
            // The tag slug where this authorization applies
            ->in('auto-gandalf')
            // What the Ai can do, full list is in the documentation/readme.
            ->can(
                replyToPosts: true,
                respondToMentions: true
            )
            // Conclude this autorization
            ->activate(),
];

Now "gandalf" replies to every post (ignoring ->in ...) if I create the user first.

Dashboard:
Flarum 1.8.5
PHP 8.2.20
MySQL 11.4.2-MariaDB

luceos

tgrothe Does the Ai user have to exist before the extension can be activated?

Yes.

tgrothe Now "gandalf" replies to every post (ignoring ->in ...) if I create the user first.

I can look into it, not sure there's a bug anywhere, but that might be possible.

xiaoxiaoi246

Hello, I don't know why this problem occurs during use. My api call is normal and the message has been returned, but I don't know why the robot in the forum does not reply to the message.
[2024-10-27 08:25:14][J7ywrQsvhbDFOMGexGTXf4u3wsiKplwa] Processing: Flarum\Notification\Job\SendNotificationsJob
[2024-10-27 08:25:14][J7ywrQsvhbDFOMGexGTXf4u3wsiKplwa] Processed: Flarum\Notification\Job\SendNotificationsJob
[2024-10-27 08:26:42][X2cKKsjUZ7dnnb6G46KqRkItxl4vz4cf] Processing: Flarum\Mentions\Job\SendMentionsNotificationsJob
[2024-10-27 08:26:42][X2cKKsjUZ7dnnb6G46KqRkItxl4vz4cf] Processed: Flarum\Mentions\Job\SendMentionsNotificationsJob
[2024-10-27 08:26:42][3IqAoRL5Srv2h4rXK75sHrJ3CutzBoSB] Processing: Flarum\Subscriptions\Job\SendReplyNotification
[2024-10-27 08:26:43][3IqAoRL5Srv2h4rXK75sHrJ3CutzBoSB] Processed: Flarum\Subscriptions\Job\SendReplyNotification
[2024-10-27 08:26:43][9MKOpwsFuVetdk0MtWWBwaBRj0K2jJzY] Processing: Flarum\Ai\Content\InteractionJob
[2024-10-27 08:26:45][9MKOpwsFuVetdk0MtWWBwaBRj0K2jJzY] Failed: Flarum\Ai\Content\InteractionJob
And the log time is not consistent with my server time
luceos

« Previous Page Next Page »