ConvOllama is a Python application that facilitates AI-driven conversations using the Ollama API. It creates engaging, multi-participant discussions on user-specified topics, with the ability to assign unique profiles to each AI participant.
- Dynamic topic generation based on user input or keywords
- Multi-participant conversations with AI-generated or user-defined profiles
- User-selectable LLM models for participants
- Configurable conversation parameters (number of participants, rounds, etc.)
- Real-time conversation display with colorized output
- Automatic saving of conversations in both JSON and Markdown formats
- Graceful handling of program interruption
- Configurable conversation history limit and logging level
- Python 3.7+
- Ollama API access
-
Clone this repository:
git clone https://github.com/rb81/convollama.git cd convollama -
Install the required dependencies:
pip install -r requirements.txt -
Ensure you have Ollama set up and running on your system.
Create a config.yaml file in the project root directory with the following structure:
moderator_model: "wizardlm2:latest"
ollama_host: "http://localhost:11434"
save_path: "~/Downloads/ConvOllama"
available_models:
- "llama3.1:latest"
- "gemma2:latest"
- "falcon2:latest"
history_limit: 3 # Keep the last 3 rounds of messages
log_level: "INFO" # Can be DEBUG, INFO, WARNING, ERROR, or CRITICALAdjust the values according to your preferences and setup.
Run the application using the following command:
python main.py
or specify a custom configuration file:
python main.py -c path/to/your/config.yaml
You will be guided through the following steps:
- Choose the number of participants (minimum 2)
- Select the LLM model for the participants
- Determine the conversation topic:
- Enter keywords and have the Moderator generate a topic
- Enter a discussion topic yourself
- For each participant:
- Provide a profile yourself
- Have the Moderator generate a profile
- Choose no profile
- Set the number of conversation rounds
The conversation will then begin. To end the conversation early, press Ctrl+C. The program will save the conversation and exit gracefully.
Conversations are saved in two formats:
- JSON file for easy parsing and analysis
- Markdown file for human-readable documentation
Both files are saved in the directory specified by save_path in the configuration.
The Moderator works well with wizardlm2. Recommended Participant models include Meta's new llama3.1, gemma2, and llama3. You can customize the list of available models in the configuration file.
This project is licensed under the MIT License - see the LICENSE file for details.
ai.collaboratedwith.me in creating this project.