Simple chat script that connects to cohere's api.
Check .Install.bat contents so you know what it does. And then run it.
This will install python module UV and build a venv in cwd, and install other modules into the venv.
If you know what you're doing: Create the env/venv you want to use, install setuptools wheel cohere to the env/venv.
Input your proxy adress by changing set HTTP_PROXY in .start-ai.bat
or
Comment out set HTTP_PROXY in .start-ai.bat.
Set and save API keys there for easy management.
Drop _system.txt files into system folder. See Jailbreak_system.txt, Aina_system.txt and Aria_system.txt for different formats that work, plaintext, json & compressed.
Also take note of structure, the files have 3 sections, it's system_prompt from start ( starts reading doc as system_prompt, no section needed ) to ai_greetings: section ( text to appear when bot is started. Not part of prompt, just for sake of copy pasting start message.), and then keys_files: section ( .txt files with keywords to inject data to ai at mention of a keyword. one file per line under keys_files: ) see example files.
Keywords files can be anywhere. Just input one filepath per row under keys_files: in Name_system.txt. No quotes.
Format of keywords file.txt:
title;word,word2,word3;Information to inject to ai.
title2;word2,word4,word5;Information to inject to ai.
ex: Capital;city,capital,Metropolis;Some information about capital that bot should know if city, capital or metropolis is mentioned.
- TTS: The tts command enables tts, but you need to host your own, and input info in the tts definition in the py file to use it.
- Retry: Removes last message and tells api to retry it to regenerate it, use retry: with colon to send a system prompt to tell the ai how to behave.
- Reset: Resets chat to start. Saves history.
- Recap: Sends a message to pause RP and recap events. This is to break AI out of loops and bad behavior. Also so you know the ai isn't confused.
- Exit: Quit, history is saved and resumed when you return. Reset history with reset command.
- Changed folder structure a little.
- Added a python file that can be run from cmd with "x:\Path\cwd\.VENV\Scripts\python.exe x:\Path\cwd\.start-ai.py" or by activating venv and running "python .start-ai.py" from cwd.
- Keyword identifier changed to title instead of first keyword.
- "retry: instructions" command is fixed.
- keywords files are fixed
Currently keywords are injected one per message, so if you trigger multiple keywords they are sent to a queue and injected one by one beside your normal messages.