Mods is a terminal-based AI client that sends prompts to large language models and streams responses back to the command line. It functions as a multi-provider AI gateway, routing queries to OpenAI, Cohere, Groq, Gemini, and local endpoints, and includes a conversation history manager that saves, caches, branches, and resumes text-based interactions. The tool also operates as a Model Context Protocol client, connecting to external MCP servers via stdio, SSE, or HTTP to extend model capabilities with specialized tools and data.
The project distinguishes itself through a config-driven provider router that selects AI providers and models at runtime based on a YAML configuration file, and a system prompt persona manager that loads named role definitions to shape model behavior. It includes automatic retry with backoff for rate-limit and transient server errors, a file-based conversation store for persisting multi-turn dialogues, and a shell pipe input adapter that accepts piped stdin, file URLs, or direct arguments as prompt input. A streaming JSON-RPC gateway handles authentication, rate limiting, and response parsing for multiple backends, while a Markdown response formatter instructs the LLM to return structured output for terminal display.
The broader capability surface includes conversation management via CLI commands to list, show, continue, and delete saved interactions, as well as interactive prompt editing and the ability to format command output with AI. Users can configure generation parameters like model, temperature, and max tokens, define custom system prompts and personas, and manage conversation history with local storage. The tool also supports loading content from URLs or local files, streaming responses incrementally, and applying custom prompt prefixes to steer model behavior.
Configuration is managed through a YAML file that can be edited directly, with settings for provider and model selection, generation parameters, and system prompt definitions.