Introduction: The Asynchronous Agentic System Operator

Welcome to the seekr documentation.

seekr is a high-performance, background-first AI Agent Manager designed to transform your terminal into an autonomous system operator. Originally built as a DeepSeek-native client, seekr has evolved into a full-featured agentic platform that runs persistently in the background, executing long-running tasks while you work elsewhere.

seekr brings "Jarvis-style" persistence to Linux via a client-server architecture, featuring a rugged Terminal UI (TUI) and a headless background daemon (seekrd) that communicates over a high-speed SSE (Server-Sent Events) interface.

Why seekr?

  • Persistent Daemon Architecture: Launch the Seekr daemon to maintain a peristent agentic presence. Close the TUI anytime; your tasks continue to execute in the background.
  • Asynchronous Task Execution: Long-running shell commands, web scrapes, and file edits continue even when you detach from the session.
  • Native Speed: Written in Rust for minimal overhead and maximum performance.
  • DeepSeek Optimized: Built as a DeepSeek original, highly optimized for DeepSeek's latest reasoning models.
  • Local Control: All agent activities happen in your terminal, under your supervision.
  • Rich TUI: A sophisticated Terminal User Interface that doesn't sacrifice power for aesthetics.

Core Concepts

The Asynchronous Daemon (seekrd)

At the heart of seekr is the background daemon. When you start seekr, it automatically stands up a headless server that manages the agent's state, tool executions, and model communication. The TUI connects to this server via SSE (Server-Sent Events) to provide a real-time window into the agent's activities.

The Agent Loop

seekr runs a continuous loop where the agent:

  1. Receives a objective or task.
  2. Analyzes the current state.
  3. Decides which tools to use. Parallel Execution: seekr can execute multiple independent tool calls simultaneously in a single turn.
  4. Executes the tools and observes the outcome.
  5. Repeats until the task is complete.

Explore the Docs

Next: Getting Started