Legacy Cleverbug Source Code

I decided to partially open-source the legacy Cleverbug code. You can find its entirety contained within cleverbug-alpha/ of this website.
To view the code on GitHub, visit this link: GitHub

The code is a mess and poorly documented. I will attempt to break it down below.

---

This version of cleverbug was split into three parts. One part ran serverside, while the other ran clientside, and the third part was a Google Spreadsheet. Yes, I used Google Sheets as a free SQL database.

Client-side code

The way users interacted with the bot was through a webpage. The webpage presented the user a simple interface: a large text area for the session's chat history, a text field for typing messages to the bot, and buttons to send the message and toggle the settings. In the settings menu were two text fields for changing your's or the bot's display name. A "think for me" button adorned the info box alongside the settings button.

When the user pressed enter or clicked the say button, the responder engine executed. It actually ran the responder clientside. The responder queried the google spreadsheet recursively with different a different SELECT call until it got a satisfactory answer. It would then add the message to the chat history. The client did not stop the user from sending messages before it had finished responding, making spam possible.

Think for me was nothing special. It simply called the responder with an empty string, effectively guarenteeing the base case of a random answer. It then placed that answer inside the user's say field.

The user posting a message to the chat also triggered the data collection, or "learning," aspect of this version. It queried my server with a set of URL arguments containg what the user wrote and other information. This was how the bot learned. It gathered sentences from the users that interacted with it and added them to the repertoire of phrases it could use. It did not generate unique phrases.

You can see the original, working client running here.

Server-side code

The server-side code was responsible for the learning aspect of the bot. I wrote the serverside in php and the clientside in HTML, CSS, and JavaScript. The server-side code has been deactivated, but is preserved here. There are two versions of the serverside code preserved here. Version one was hard coded, save for the client-secret file needed for Google Sheets authentication. Version two loaded a file named cleverbug_badwords.txt which contained a comma-separated list of tokens. I moved the bad word detection to a separate file so that I could easily add new tokens without fear of accidentally breaking the php code.

When the server recieved a request from the client, it first checked the body for blacklisted words as described by cleverbug_badwords.txt. If a bad word was found, the server rejected the request. This step helped prevent (and worked surprisingly well) the bot from learning racist or offensive phrases. The server also performed basic duplicate-detection by checking the input string as lowercase with the spaces and double letters removed against a column in the spreadsheet which stored all the phrases lowercase without spaces. The spreadsheet calculated those automatically via spreadsheet ArrayFormula, so the server simply ran a select query on that column. Like the bad word detection, this duplicate detection worked surprisingly well.

If the code did not find a bad word or a duplicate phrase (and it checked for variations), then it proceeded to the next step of adding it to the database. Here, it initialized the Google Spreadsheet API library for php (verison 2.2.0) and authenticated using the credentials described in client_secret.json.

After authentication, it called the Spreadsheet API to add a row to the spreadsheet. The row, once added, was instantly available to the clients, enabling real-time learning.

Two files, cleverbot_up.php and cleverbot_respond.php also exist. These files contained a single instruction which was to include a different php file. The clients called these two files. The reason I did this was because I didn't know anything about web security and I stored my credentials in plain text, albeit outside of public_html. I believed at the time that by storing the php files that did the actual work outside of public_html, I would prevent hackers from tricking my server into downloading the php code and seeing how it worked. Of course, web security doesn't actually work that way.

Conclusion

I didn't want to delete this code, so I decided to make it public, not that this code is worth anything. If you want to read it, go ahead. This code turned into a Discord bot with the same name, which you can invite to your server by visiting the working client (linked above) and using the link in the footer.