Styx

Product Introduction

1. Definition: Styx is an Agentic Development Environment (ADE) for macOS, designed to centralize and secure the workflow of developers using AI coding assistants. It is a desktop application that consolidates multiple code repositories, AI agents (like Claude Code, Codex, Gemini, and Cursor), and their required infrastructure access into a single, controlled interface.
2. Core Value Proposition: Styx exists to solve the fragmentation and security risks inherent in modern, AI-assisted software development. Its primary value is providing a unified workspace where developers can manage all AI agent activity and securely delegate temporary, auditable access to production systems, ensuring nothing reaches production without explicit human approval.

Main Features

1. Unified Project & Agent Dashboard: Styx provides a single-window overview of all active projects (repositories), each with its associated AI agents and their current branches. The dashboard uses a visual board (e.g., ASBLINCXSA+) to show project status, active agents, and pending approvals at a glance. Technically, it monitors local git repositories and agent processes, aggregating their states into a real-time status UI.
2. Secure, Temporary Access Grants (The "Crossing"): This is the core security mechanism. When an AI agent attempts an action requiring credentials (e.g., a database migration, cloud deploy, or git push), Styx intercepts the request. The agent is paused, and the user is prompted via notification, dock badge, or in-app alert. The user can grant scoped (read/write), time-bound (e.g., 1 hour, session-based) access, often requiring biometric authentication (Touch ID) for production actions. Credentials are stored solely in the system's secure enclave (macOS Keychain), never exposed to the agents.
3. Integrated Multi-Agent Workspace with Change Control: Styx allows multiple AI agents (Claude, Codex, Gemini, Cursor, shell) to work concurrently on different branches of the same project. Each agent's changes are tracked per "turn" and can be reviewed in a unified diff viewer. Users can revert individual turns. Features like "Land" automate merging with pre-merge checks, and conflict resolution is pushed back to the originating agent.
4. Live Preview & Tech Debt Audit: The "Design" tab renders a live, local preview of web applications (desktop/tablet/phone views) alongside the code. The "Tech Debt Audit" feature uses an AI agent to statically analyze the codebase, identifying discrepancies (e.g., outdated README instructions), anti-patterns, and potential bugs, reporting them with file and line references without making changes.
5. Comprehensive Audit Logging & Policy Engine: All access requests, grants, and agent actions are written to an immutable audit log. A configurable policy engine allows rules like "Auto-approve reads on staging" or "Always ask + MFA for prod writes." This creates a non-repudiable trail of all AI-agent interactions with sensitive systems.

Problems Solved

1. Pain Point: Context Switching and Tool Sprawl. Developers juggle multiple terminals, IDE windows, browser tabs for different AI assistants, and cloud consoles. Styx consolidates this into one application, reducing cognitive load and tab overload.
2. Pain Point: Uncontrolled AI Agent Access. A major risk of AI coding assistants is their potential to execute commands with broad, persistent credentials. Styx eliminates this by enforcing a zero-trust model where agents have no inherent credentials and must request temporary, audited passes for specific actions.
3. Target Audience: Full-Stack and DevOps Engineers who regularly use AI coding tools (Claude Code, Cursor) and manage deployments across platforms like Vercel, AWS, GCP, and Supabase. It's also critical for Tech Leads and Engineering Managers who need oversight and audit trails for AI-assisted work in their teams.
4. Use Cases:
   • Safe Database Migrations: An AI agent (Codex) writes and tests a migration locally, but the supabase db push --linked command to production is intercepted. The engineer reviews the SQL and grants a one-time, write-scoped pass.
   • Supervised Production Deployments: An agent prepares a release and runs vercel deploy --prod. Styx pauses the process, requests Touch ID, and only then executes the deploy, logging everything.
   • Multi-Agent Feature Development: A developer uses Claude to refactor a checkout module on branch fix/checkout while simultaneously having Gemini update documentation on docs/readme, all monitored from one board with separate, isolated workspaces.

Unique Advantages

1. Differentiation: Unlike pure AI code editors (Cursor, Windsurf) or standalone CLI agents, Styx is a control plane. It doesn't replace your IDE or terminal but sits above them, orchestrating and securing interactions between multiple AI tools and your infrastructure. Compared to manual credential management, it provides granular, just-in-time access instead of broad, static keys.
2. Key Innovation: The "Crossing" metaphor and its technical implementation. The product's core innovation is the interceptor layer that sits between AI agent commands and the external APIs/services they call. This allows for mandatory human-in-the-loop approval for any action crossing into live environments, coupled with ephemeral credential injection—a significant architectural shift for AI development safety.

Frequently Asked Questions (FAQ)

1. How does Styx handle my API keys and passwords securely? Styx never stores credentials within its own database. All secrets (e.g., GitHub tokens, cloud keys) are stored exclusively in your operating system's secure storage: the macOS Keychain or Windows Credential Manager. Agents request access, and Styx temporarily injects scoped tokens into their environment, which are automatically revoked.
2. Can I use Styx with my existing IDE like VS Code or JetBrains? Yes. Styx has a built-in editor for reviewing agent changes, but it is designed to integrate with your primary IDE. It can detect and import settings (themes, keybindings, recent projects) from VS Code, Cursor, Windsurf, JetBrains IDEs (WebStorm, IntelliJ), and provides "Open in..." functionality for Zed and Neovim.
3. What happens if two AI agents modify the same file? Styx includes conflict awareness. Agents are warned if another agent is working on the same file. If a conflict occurs during a merge (e.g., when landing a branch), the conflict is passed back to the original agent to resolve, maintaining the principle that the agent responsible for the changes handles the integration issues.
4. Is there a way to automate approvals for low-risk actions? Yes. Styx includes a configurable policy engine. You can create rules to auto-approve certain actions, such as "Always allow read access to GitHub" or "Auto-approve deployments to preview environments," while still requiring manual approval for production writes.
5. Does Styx work with on-premises or self-hosted services like GitLab or private servers? The product explicitly supports GitHub, Vercel, AWS, Google Cloud, and Supabase via OAuth/API keys. For other targets like private servers, it supports SSH connections using the existing identities (SSH keys) on your computer, applying the same request-and-approval model to commands run over SSH.