Summary
Several declared viewpoints on one question, instead of one averaged answer.
The traits are stored, not retyped: an agent defined once by first principles, systems thinking and risk management answers in that character every time you call it. Every answer goes through DeepSeek, so that key is where the cost lands.
What it is
A Go server that creates agents from named personality traits and simulates their answers through the DeepSeek API. Five tools cover creation, answering and management.
What you get
- An agent created from a name and a list of core traits — expert_personality_generation
- That agent's answer to a question, with the planned round count and the current round passed on each call — agent_answer
- One agent fetched, all agents listed, one deleted — get_agent, list_agents, delete_agent
- Multi-round dialogue, where the caller signals whether more rounds are needed
- A base for multi-agent research along the lines of a collaborative research project
Requirements
Go 1.24.1 or higher to build, and a DeepSeek API key in DEEPSEEK_API_KEY. Clone the repository and run make build, then point your client at the binary. LOG_LEVEL, LOG_PATH, CONFIG_PATH, PORT and DEBUG are optional.