# Bestie

> A native terminal coding agent harness that gives you total visibility into every action the AI takes, with cross-platform sandboxing and support for multiple LLM providers.

Bestie is a native terminal coding agent harness written in Dart, built by jolexxa and licensed under the Mozilla Public License 2.0. It runs natively on Windows x64, Apple Silicon, and Linux x64, and is self-described as being in alpha. The project's stated goal is to give users total power over AI coding agents — nothing happens without you seeing it.

## What It Is

Bestie is a terminal-based coding agent harness: a local application that wraps AI coding agents and exposes every input, output, and action through a detailed UI. Unlike cloud-based coding assistants, Bestie runs entirely on your machine, sandboxes agent activity, and lets you inspect what the agent is doing at every step. It connects to external LLM providers (Fireworks.ai, OpenRouter.ai, and any OpenAI-compatible server) rather than bundling its own model.

## Visibility and Control Architecture

The core design principle is transparency. Bestie provides a details pane showing inputs and outputs for every agent action, a command palette, and a visual options menu so users never have to edit raw JSON config files. Key control features include:

- **Cross-platform sandbox**: wide read access but limited writes scoped to the working directory
- **Configurable network access levels**: control what the agent can reach on the internet
- **Session save and load**: resume agent sessions across restarts
- **Internet access via curl-impersonate**: reduces the likelihood of the agent being blocked by CAPTCHAs during web requests

## Agent Capabilities

On the agent side, Bestie supports a range of agentic primitives:

- Background jobs and agentic terminal shells
- Subagents for delegating subtasks
- Automatic and manual conversation compaction
- Parallel tool execution via thread pools
- A pull-based paradigm for handling long outputs and shell processes
- A sandboxed bash environment available on every supported OS

## Platform and Technology

Bestie is written in Dart, chosen for its native compilation, type safety, sound null safety, and strong FFI support for integrating with native OS libraries. The README notes that pseudo-terminals, clipboards, process spawning, and sandboxes all behave differently across Windows, macOS, and Linux, and Bestie implements platform-specific abstractions for each. Dart isolates are used extensively for concurrency without low-level locking concerns. The app also includes a built-in terminal emulator/multiplexer, themes, mascot characters, an ASCII math renderer, and mouse support.

## Roadmap and Current Status

The project launched its first release (v0.1.0) in September 2026 and is explicitly described as alpha-quality software. The roadmap includes planned support for the Agent Client Protocol (ACP), Model Context Protocol (MCP), skills, agents.md, localization, inline skill references, and plans/todos. The repository is open for issues, contributions, and community feedback via a Discord server.

## Features
- Details pane showing inputs/outputs for every agent action
- Cross-platform sandbox with wide read access and limited writes
- Configurable network access levels
- Command palette
- Session save and load
- Support for Fireworks.ai, OpenRouter.ai, and OpenAI-compatible servers
- Built-in terminal emulator and multiplexer
- Background jobs
- Agentic terminal shells
- Subagents
- Automatic and manual conversation compaction
- Sandboxed bash environment on all platforms
- Parallel tool execution via thread pools
- Pull-based paradigm for long outputs and shell processes
- Internet access via curl-impersonate
- Themes and mascot characters
- ASCII math renderer
- Visual options menu
- Mouse support
- Terminal and markdown text reflowing with scroll position preservation

## Integrations
Fireworks.ai, OpenRouter.ai, OpenAI-compatible servers, curl-impersonate

## Platforms
WINDOWS, MACOS, LINUX, API, CLI

## Pricing
Open Source

## Version
v0.1.0

## Links
- Website: https://github.com/jolexxa/bestie
- Documentation: https://github.com/jolexxa/bestie
- Repository: https://github.com/jolexxa/bestie
- EveryDev.ai: https://www.everydev.ai/tools/bestie