This project consists of Agent Server and Trigger Server. The Agent Server stores system prompts, configures LLM models, and defines execution roles. The Trigger Server binds a workspace (local directory), connects to the Agent Server, automatically downloads tools, and invokes the LLM through loaded roles. A Trigger Server can connect to multiple Agent Servers and load their roles. While loading roles, any associated tools are automatically downloaded. This project is built by VibeCoding.
- models: Configuration hub for LLMs, including OpenAI chat completions and Anthropic APIs.
- components: Dynamic prompt templates, including static and embedded types. Embedded types inject workspace environment variables into prompt templates using the
{env_name}syntax. - tools: Executable scripts or binary files, including schemas, descriptions, and dependency configurations.
- roles: Bundles prompt components, model configurations, and tools to define a role, acting as a session template for invoking the LLM.
- workspace: The designated working folder, utilizing workspace files and environment context.
- servers: Gateway to connect to Agent Servers and select which roles to load.
- events: LLM invocation tasks that use workspace and environment variables to construct context for LLM completions.
- triggers: Binds local files or directories to events. Supports three execution modes: direct ("call once"), periodic ("call at specified intervals"), and edit ("call when files are modified").
- envs: Workspace environment variables, mapping local files as environment keys in the format
file:filename.
- Create components, tools, and model configurations.
- Define and configure roles.
- Start the server.
- Create or bind a workspace.
- Connect to the Agent Server, select roles, and download their associated tools.
- Start the daemon process.
- Define execution events.
- Create a trigger to bind files or folders to an event.