Cybara Launches as Open-Source Agent Operating System for Developers

Cybara, a new open-source AI agent platform, has launched with a focus on giving developers and operators a single stack for building and running agents that can code, automate browsers, manage messaging workflows, and execute wallet operations. The platform is self-hosted, built on Bun, and ships with a web UI, CLI, desktop shells for Windows, macOS, and Linux, a mobile companion app, and a broad tool layer.

What Cybara Offers

At its core, Cybara is an agent operating system designed to plan, execute, verify, and report while keeping the operator in control. It supports multiple interfaces:

• Web and Tauri UI with persisted workspaces, live activity, embedded previews, and agent selection.
• Terminal chat with session management, queueing, steering, tool approvals, and slash commands.
• Desktop shells — both Tauri and native macOS (SwiftUI with AppKit) — over the same Bun runtime.
• Mobile app (iOS/Android) for remote session management, connected via QR pairing with revocable tokens.

The tool layer includes 80+ built-in tools spanning code analysis, file manipulation, image/PDF processing, web scraping, browser automation, communication adapters (Discord, Slack, Telegram, email, SMS, WhatsApp), and on-chain operations via encrypted local wallet controls.

Developer-Focused Architecture

Cybara's runtime is designed for extensibility and control. It uses a dynamic prompt assembly that builds each model request from session state, tool context, provider metadata, and operator policies. Tool policy and sandbox enforcement remain authoritative even if prompt text or untrusted content tries to override them.

Key technical highlights:

• Provider support: OpenAI, Anthropic, Google Gemini, xAI Grok, DeepSeek, and local runtimes like Ollama and vLLM, with dynamic model discovery.
• Multi-key credential pools that rotate automatically on rate-limit/auth errors.
• Anthropic prompt caching applied to every Claude request for ~75% input-token savings on multi-turn sessions.
• MCP support: Cybara both consumes external MCP servers and exposes its own tools as an MCP server.
• Plugin system: Bundled, local, and workspace plugins with a Settings UI and CLI management.
• REST API and streaming interfaces for external integration.

Getting Started

Installation is straightforward via a single command on macOS/Linux:

curl -fsSL https://cybara.ai/install.sh | sh

On Windows PowerShell:

irm https://cybara.ai/install.ps1 | iex

Then launch with cybara start or run without installation via bunx cybara.

For production deployment, Cybara provides a Docker option:

docker pull cybara/cybara
docker run -d -p 4269:4269 cybara/cybara

License and Community

Cybara is released under the MIT license and is available on GitHub. The project encourages contributions: development is done on the dev branch, with PRs directed there. The main branch is reserved for releases.

Developers interested in self-hosting their agent infrastructure will find Cybara a compelling option, especially those needing strong operator controls, multi-channel support, and on-chain execution capabilities.