________ __
/_ __/ /_ ___ _________ / /____
/ / / __ \/ _ \/ ___/ __ \/ / ___/
/ / / / / / __/ / / /_/ / (__ )
/_/ /_/ /_/\___/_/ \____/_/____/
/\
/ \
/ /\ \
/ / \ \
/_/ / /
/ /
/ /
/ /
/ /
/_/
Zeus is a fast, local coding agent for your terminal. It's a community-driven fork of openai/codex focused on real developer ergonomics: Browser integration, multi-agents, theming, and reasoning control — all while staying compatible with upstream.
Interactions with git take fucking ages, no idea why they're so slow, should maybe automatically swap to a smaller model for that
- Auto Drive upgraded – hand
/autoa task and it now plans, coordinates agents, reruns checks, and recovers from hiccups without babysitting. - Unified settings –
/settingscentralizes limits, model routing, themes, and CLI integrations so you can audit configuration in one place. - Card-based activity – Agents, browser sessions, web search, and Auto Drive render as compact cards with drill-down overlays for full logs.
- Turbocharged performance – History rendering and streaming were optimized to stay smooth even during long multi-agent sessions.
- Smarter agents – Mix and match orchestrator CLIs (Claude, Gemini, GPT-5, Qwen, and more) per
/plan,/code, or/solverun.
Read the full notes in release-notes/RELEASE_NOTES.md.
- 🚀 Auto Drive orchestration – Multi-agent automation that now self-heals and ships complete tasks.
- 🌐 Browser Integration – CDP support, headless browsing, screenshots captured inline.
- 🤖 Multi-agent commands –
/plan,/codeand/solvecoordinate multiple CLI agents. - 🧭 Unified settings hub –
/settingsoverlay for limits, theming, approvals, and provider wiring. - 🎨 Theme system – Switch between accessible presets, customize accents, and preview live via
/themes. - 🔌 MCP support – Extend with filesystem, DBs, APIs, or your own tools.
- 🔒 Safety modes – Read-only, approvals, and workspace sandboxing.
Zeus is part of an evolving ecosystem of AI coding agents. Here's how it compares to other tools:
- Aider – Git-aware AI pair programmer with code editing capabilities
- OpenCode – Computer use-enabled coding agent
- Cline – VS Code extension for agentic coding workflows
- Goose – Extensible coding agent with custom tool support
- Cursor – VS Code-based IDE with integrated Claude autocomplete and chat
- Anthropic's Claude – Web-based chat with programming capabilities
- GitHub Copilot – IDE plugin for code completion (no agentic reasoning)
- LangGraph – LLM frameworks for agentic workflows with state management
- SWE-agent – Automated agent for software engineering tasks
- Qdrant Supervisor – Multi-agent framework for autonomous systems
- Serena – Semantic code analysis toolkit (planned Zeus integration)
- Brokk – Code understanding and analysis platform
- Octocode - Octocode for code/semantic understanding
Zeus demonstrates key patterns in modern AI agent design:
- Multi-model support – Switch between Claude, Gemini, GPT-5, Qwen without rewriting
- MCP integration – Standardized protocol for extending agent capabilities
- Browser automation – Screenshot + navigation for web-based workflows
- Multi-agent orchestration – Coordinate specialized agents for complex tasks
- Session persistence – Memory and context across conversations
- Rich output – Streaming, tables, formatted code, terminal UI
- ✅ CLI chat interface with terminal UI
- ✅ Multi-model support (Claude, GPT-4, Gemini, Qwen)
- ✅ Browser integration (CDP)
- ✅ Basic MCP support
- ✅ Read-only and approval modes
- ✅ Auto Drive orchestration
- ✅ Settings hub
Smart context management
- Relevant code caching for faster responses
- Semantic search across project files
- Intelligent token counting and model routing
- Adaptive context window management
Observability Comprehensive trace capture and visualization for agent execution flows:
- Full request/response logging with structured formatting
- Performance metrics (latency, token usage, cost, concurrency)
- Agent execution DAG visualization
- Error tracking and categorization by severity
- Export traces for external analysis or compliance reporting
Implementation approach:
- OpenTelemetry instrumentation throughout Zeus for standardized trace collection
- Optional backend configuration (SigNoz, SkyWalking, OneUptime, or local)
- Inline trace inspection via
/tracescommand with filters and search - Automatic trace linking between parent and child agent tasks
Multi-session context management
- Cross-session memory and learning
- Persistent agent personas
- Collaborative workflows (multiple team members)
Agentic patterns
- Specialized micro-agents for specific domains
- Reflection and self-correction patterns
- Multi-turn planning and refinement
Enterprise features
- Audit logging and compliance hooks
- Custom model provider support
- Advanced permission and workspace management
npx -y @just-every/zeusnpm install -g @just-every/zeus
zeus // or `zeusr` if you're using VS CodeNote: If another tool already provides a zeus command (e.g. VS Code), our CLI is also installed as zeusr. Use zeusr to avoid conflicts.
Authenticate (one of the following):
- Sign in with ChatGPT (Plus/Pro/Team; uses models available to your plan)
- Run
zeusand pick "Sign in with ChatGPT"
- Run
- API key (usage-based)
- Set
export OPENAI_API_KEY=xyzand runzeus
- Set
Zeus supports orchestrating other AI CLI tools. Install these and config to use alongside Zeus.
# Ensure Node.js 20+ is available locally (installs into ~/.n)
n install 20
# Install Claude CLI (requires Anthropic API key)
npm install -g @anthropic-ai/claude-cli
claude --version
# Install Gemini CLI
npm install -g @google/generative-ai-cli
gemini --version
# Install Qwen CLI (requires Alibaba Cloud API key)
npm install -g @alibaba-cloud/qwen-cli
qwen --version# Connect Zeus to external Chrome browser (running CDP)
/chrome # Connect with auto-detect port
/chrome 9222 # Connect to specific port
# Browser sessions are available in command selection dropdown or `/solve` when working with web apps
zeus [options] [prompt]
Options:
--model <name> Override the model (gpt-5, claude-opus, etc.)
--no-approval Auto-approve file changes (caution!)
--read-only Prevent modifications (analysis only)
--auto Start in Auto Drive mode
--help Show help message
Zeus can remember context across sessions:
- Create an
AGENTS.mdorCLAUDE.mdfile in your project root:
# Your project name
This is a React TypeScript application with:
- `/client/` - React components and UI
- `/server/` - Backend services- Session memory: Zeus maintains conversation history
- Codebase analysis: Automatically understands project structure
For automation and CI/CD:
# Run a specific task
zeus --no-approval "run tests and fix any failures"
# Generate reports
zeus --read-only "analyze code quality and generate report"
# Batch processing
zeus --config output_format=json "list all TODO comments"
Zeus supports MCP for extended capabilities:
- File operations: Advanced file system access
- Database connections: Query and modify databases
- API integrations: Connect to external services
- Custom tools: Build your own extensions
Configure MCP in ~/.zeus/config.toml Define each server under a named table like [mcp_servers.<name>] (this maps to the JSON mcpServers object used by other clients):
[mcp_servers.filesystem]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/project"]
Configuration is stored in ~/.zeus/config.toml:
[general]
default_model = "claude-opus"
approval_mode = "review" # "review", "auto", or "manual"
[themes]
current = "dark"
accent_color = "blue"
[mcp_servers]
# Your MCP server configurations here
- TUI – Terminal user interface (Rust using Ratatui)
- Core Agent – Multi-model orchestration
- Tool Runner – File operations, browser control, MCP
- Model Provider – Abstraction for LLM APIs
- Browser integration – Chrome DevTools Protocol (CDP) for automation
- Streaming – Real-time token streaming from all supported LLMs
- Stateful UI – Maintains context across tool executions
- MCP – Pluggable tool system for extending capabilities
# Clone the repo
git clone https://github.com/just-every/zeus.git
cd zeus
# Install dependencies
npm install
# Run in development
npm run dev
# Run tests
npm test# Build for your platform
npm run build
# Build for all platforms (requires platform-specific tools)
npm run build:all
Zeus's design and roadmap are informed by proven patterns from leading open-source projects:
- ModelScope ms-agent – Agent-centric orchestration with async/await patterns, chain-based multi-agent workflows, pluggable tool/plugin system, and configuration-driven lifecycle management. Demonstrates clear callback-based observability patterns.
- AutoGen – Multi-agent conversations, role-based agents, and complex conversation flows with human-in-the-loop patterns. Reference for orchestration state management.
- Smolagents – Minimal agent framework with composable tools and LLM-based planning. Shows alternative to hierarchical agent design.
- Charm Bubble Tea – Elegant TUI framework (Go) inspiring our streaming and state model
- Hurl – CLI tool with great UX design patterns
- GitHub CLI – Example of well-structured command parsing and output formatting
- Playwright – Cross-browser automation reference; demonstrates resilient navigation patterns
- Puppeteer – Chrome DevTools Protocol patterns and high-level API design
- Nix – Declarative system configuration patterns
- Just – Command runner with clear, readable syntax; influenced our
/command design
- Anthropic's Streaming – Streaming response patterns for real-time user feedback
- Ollama – Local model management and API reference design
- Mozilla's Content Security Policy – Approval sandboxing patterns
- Temporal Workflows – Durable execution and approval workflow patterns
Contributions are welcome! Areas we're actively developing:
- New model provider integrations
- MCP server implementations
- TUI improvements and themes
- Documentation and examples
- Bug fixes and performance
MIT – See LICENSE file