A simple console application to ask multiple large language models for a consensus answer using openrouter.ai.
dotnet build
Run the application with:
dotnet run --project src/Consensus.Console/Consensus.Console -- "your question here"
The program will prompt you for models and your OpenRouter API key if not provided in the OPENROUTER_API_KEY environment variable.
You can optionally create a markdown log file of each model's response. When prompted, choose Minimal to log short summaries of each model's changes or Full to store every response separated by ----------- lines. At the end of the log file the final summary from <ChangesSummary> tags is also collected for quick reference and notes which model made each change.
By default the log and response files use the first ten characters of the prompt for their names and are overwritten on each run. Set the CONSENSUS_UNIQUE_FILES environment variable to create new files with a timestamp instead.
The system prompts used for each model are stored as text files in Resources and embedded in the application.
After all models have responded, the final model generates a summary of the changes inside <ChangesSummary> tags which credits each change to the model that suggested it and is printed to the terminal.
The Consensus.Api project exposes three endpoints. The /consensus route returns a JSON payload with the answer text, path to the generated file, its change summary and optional log path. The /consensus/stream route implements the OpenAI streaming format and emits a chunk after every LLM response plus a final result followed by data: [DONE]. Finally /log?path=PATH returns the contents of a previously generated log file.
When running the API locally, navigate to /swagger to explore the endpoints and models using Swagger UI.