evotai/evot

The lightest harness for agentic work.

What it solves

Evot is a lightweight coding agent for the terminal that allows developers to automate software engineering tasks through a short prompt. It aims to reduce the overhead (tokens and cost) associated with agentic workflows by providing a minimal harness that relies more on the model's reasoning capabilities than complex internal logic.

How it works

The agent operates within a terminal-based user interface (TUI) and uses four core tools—read, bash, edit, and write—to execute tasks. It supports coordinated execution, where shell commands can run in the background while the model continues to work on independent tasks, reducing idle time. It can be integrated with external session hooks via NDJSON to track session lifecycles (e.g., working, blocked, idle).

Who it’s for

Developers who want a low-cost, efficient coding agent that integrates directly into their terminal and supports a wide variety of LLM providers (including Anthropic, OpenAI, DeepSeek, and OpenRouter).

Highlights

  • Minimalist Design: Uses approximately 1k tokens for its harness, reducing tool call frequency and costs.
  • Asynchronous Execution: Backgrounds long-running commands to keep the model active.
  • ** uma a wide range of LLM support**: Compatible with multiple providers and supports provider-native "remote compaction" for OpenAI models.
  • External Hooks: Provides a session-adapter protocol for live status tracking in external tools.

Related

  • Project
  • Project
  • Project
  • Project