Nerve
Simplifies the development, running, evaluation, and orchestration of LLM-based agents using YAML and a CLI.
About
Nerve is a simple yet powerful Agent Development Kit (ADK) designed for technical users seeking programmable, auditable, and reproducible automation using large language models. Defining agents is streamlined through a clean YAML format, encompassing system prompts, tasks, tools, and variables within a single file. Built-in tools, along with extensibility via shell commands, Python functions, or remote tools, empower agents with fully typed and annotated capabilities. Nerve stands out with its native MCP support, functioning as both a client and server, enabling agent teams and deep orchestration. Evaluation mode allows benchmarking agents with YAML, Parquet, or folder-based test cases, ensuring reproducible tests and structured output logging. Workflows can be constructed by composing agents into linear pipelines for multi-step automations, while the foundation on LiteLLM offers LLM-agnostic support for models like OpenAI, Anthropic, and Ollama.
Key Features
- Evaluation mode for benchmarking agents.
- LLM-agnostic support through LiteLLM.
- Native MCP support (client & server).
- Built-in tools and extensibility with shell commands, Python functions, and remote tools.
- Declarative agent definition using YAML.
- 980 GitHub stars
Use Cases
- Building automated workflows using LLM-based agents.
- Orchestrating teams of agents using MCP.
- Benchmarking and evaluating agent performance.
