Batteries-included development environment for agentic workflows.
As tools like Claude Code, OpenCode and Codex gain agency, they need full system access to be effective. Running them locally carries risk (rm -rf accidents, credential leaks) and friction (installing runtimes, approving every action).
EnvHaven is an isolated, remote environment with every major AI coding tool built-in. The container provides the isolation, so all agents ship with full access by default. No approval prompts, no confirmation dialogs. Your local machine stays untouched.
Workflow Freedom:
- Browser-only: Deploy the image, open VS Code in your browser, run
opencode. Zero setup. - Local editor + remote AI: Use the Haven CLI to sync your project. Edit in Neovim, Zed, Jetbrains, et al. while agents run in the container. Changes sync in ~200ms.
🛡️ Zero-Risk AI environment Give autonomous agents full access in a containerized environment. If they break something, nuke the workspace and start over. Your local machine stays untouched.
🖥️ Pick Up Where You Left Off Start Claude Code on your laptop, close the lid, resume from your desktop: same session, same state. Your AI agents and dev servers keep running while you're away. Come back from any device.
🌲 Evergreen AI Tooling The landscape moves too fast to manage manually. EnvHaven ships with 17 AI CLI tools (Claude Code, OpenCode, Codex, Kimi Code CLI, Muse Code, Goose, etc.) built-in and ready to run.
🌐 Instant Public URLs (Managed only)
Every managed workspace gets a wildcard *.envhaven.app domain. Deploy a web app, webhook receiver, or Discord bot. Live instantly, with no ngrok, no tunnels, no DNS.
💾 Persistent & Owned Unlike ephemeral environments (Codespaces) or walled gardens (Replit), EnvHaven workspaces are persistent Linux environments you own.
⚡ Open Model
- Self-Hosted: Free forever, open-source Docker image and CLI. Run it on your own server.
- Managed (from $12/mo): Zero-config hosting with custom domains, always-on workspaces, and no DevOps.
| Persona | Use Case |
|---|---|
| Vibe Coders | That ship by prompting their way to products. Need a "magic box" where code just runs, already fully setup. |
| AI-Native Devs | Power users of agentic coding tools like Claude Code or OpenCode that want a standardized, pre-configured backend for their agents. |
| Setup Purists | Developers married to their beautifully tuned and customized editor. Use Haven CLI to keep your workflow while agents run wild in a contained environment. |
| Bot Builders | Those who vibe-code 24/7 persistent scripts (Discord, Telegram or even trading bots). |
Deploy the image and code immediately in the browser.
docker run -d \
--name envhaven \
-p 8443:8443 \
-p 2222:22 \
-e PASSWORD=password \
-e SUDO_PASSWORD=password \
ghcr.io/envhaven/envhaven:latest- Open
http://localhost:8443(Password:password) - Open terminal; you're dropped into a persistent session
- Run
opencode,claude, orcodexand start coding
Use your local editor. Files sync bidirectionally. Agents run in the container.
1. Install Haven CLI
curl -fsSL https://envhaven.com/install.sh | bash2. Connect to your workspace
cd my-project
# Managed workspace (shorthand)
haven connect . myproject-alice
# Self-hosted container
haven connect . abc@your-server.com:2222Files sync in ~200ms.
3. Run remote commands
haven opencode # Runs OpenCode in the container
haven make build # Runs make in the container
haven npm install # Runs npm in the containerSelf-hosted? You'll need to configure SSH keys first. See SSH Access.
The image (ghcr.io/envhaven/envhaven) comes batteries-included.
No API keys are pre-set. You can set them directly in the UI via the bundled EnvHaven Extension in code-server.
You can also provide your own keys via env vars (-e ANTHROPIC_API_KEY=...) or config files.
The image includes 17 AI coding agents.
| Tool | Command | Description | Authentication |
|---|---|---|---|
| Claude Code | claude |
Anthropic's official CLI | ANTHROPIC_API_KEY, CLAUDE_CODE_OAUTH_TOKEN, or /login in the CLI |
| Codex | codex |
OpenAI's coding agent | ChatGPT/device login, an API key via codex login --with-api-key, or CODEX_ACCESS_TOKEN |
| OpenCode | opencode |
AI coding agent from Anomaly | ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, or GOOGLE_API_KEY |
| Antigravity CLI | agy |
Google's agent-first CLI | GEMINI_API_KEY, or run agy to sign in |
| Grok Build | grok |
SpaceXAI's coding agent | XAI_API_KEY or grok login --device-auth |
| Kimi Code CLI | kimi |
Moonshot AI's open-source coding agent | kimi login --region global (use mainland-cn for kimi.com) |
| Muse Code | muse |
Meta's agentic coding assistant | META_API_KEY or browser sign-in on first run |
| OpenClaw | openclaw |
Personal AI assistant gateway | Provider API keys or openclaw onboard |
| Hermes Agent | hermes |
Self-improving agent from Nous Research | Provider API keys or hermes setup |
| Pi | pi |
Minimal extensible coding agent | Provider API keys or /login in the CLI |
| Goose | goose |
Extensible open-source agent | ANTHROPIC_API_KEY or OPENAI_API_KEY; run goose configure |
| Qwen Code | qwen |
Alibaba's coding assistant | OPENAI_API_KEY with an OpenAI-compatible endpoint |
| Mistral Vibe | vibe |
Mistral's minimal CLI coding agent | MISTRAL_API_KEY |
| Factory | droid |
Factory's Droid coding agent | Browser login or FACTORY_API_KEY |
| Kiro | kiro-cli |
AWS-powered AI CLI | Browser or device-flow login |
| Amp | amp |
Frontier coding agent | AMP_API_KEY or amp login |
| fx | fx |
Tiny native coding agent from Vercel Labs | AI_GATEWAY_API_KEY, or run fx to sign in |
EnvHaven workspaces are structurally isolated. The container provides the isolation, so all agents are pre-configured with full access by default. No approval prompts, no confirmation dialogs. Agents execute immediately.
You can re-enable prompts per tool by editing the relevant config or unsetting the env var. See the in-workspace AGENTS.md for details.
| Language | Version | Notes |
|---|---|---|
| Node.js | 22.x | LTS |
| Bun | 1.3.5 | Fast JS runtime |
| Python | 3.12 | Via mise |
| Go | 1.22 | |
| Rust | Stable | Via rustup |
- Core:
zsh,git,curl,wget,unzip - Utils:
ripgrep,fd,jq,sqlite3,htop - Media:
ffmpeg(video/audio),playwright(browser automation) - Integrations:
gh(GitHub CLI),docker(client)
The CLI enables a hybrid workflow: Local Editor + Remote Compute.
- Latency: ~200ms bidirectional sync.
- Compatibility: Works with any editor.
- Independence: The sync daemon runs in the background; close the CLI and sync continues.
┌────────────────────────────────────────────────────────────────────────┐
│ YOUR MACHINE │
│ │
│ ┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐ │
│ │ Local Editors │ │ Haven CLI │ │ Local Builds │ │
│ │ │ │ │ │ │ │
│ │ vim │ │ haven connect │ │ binaries │ │
│ │ vscode │ │ haven status │ │ native apps │ │
│ │ zed │ │ haven <cmd> │ │ certs+signing │ │
│ │ │ │ │ │ │ │
│ └────────┬────────┘ └────────┬────────┘ └────────┬────────┘ │
│ │ │ │ │
│ ▼ ▼ ▼ │
│ ┌──────────────────────────────────────────────────────────────────┐ │
│ │ ~/projects/myapp/ │ │
│ └──────────────────────────────────────────────────────────────────┘ │
│ ▲ │
└───────────────────────────────────┼────────────────────────────────────┘
│
Files sync in ~200ms
via SSH + Mutagen Protocol
│
┌───────────────────────────────────┼────────────────────────────────────┐
│ ▼ │
│ ┌──────────────────────────────────────────────────────────────────┐ │
│ │ /config/workspace/myapp/ │ │
│ └──────────────────────────────────────────────────────────────────┘ │
│ ▲ ▲ ▲ │
│ │ │ │ │
│ ┌────────┴────────┐ ┌────────┴────────┐ ┌────────┴────────┐ │
│ │ Claude Code │ │ OpenCode │ │ Codex │ │
│ │ Long-running │ │ Long-running │ │ Task-based │ │
│ │ AI sessions │ │ AI sessions │ │ AI sessions │ │
│ └─────────────────┘ └─────────────────┘ └─────────────────┘ │
│ │
│ ENVHAVEN CONTAINER │
└────────────────────────────────────────────────────────────────────────┘
Why this matters:
- AI agents need contained compute: They require massive context windows, long-running processes, specific toolchains, and might do dangerous stuff some time.
- You need your editor: You rely on your custom keybindings, themes, and muscle memory.
- Flexibility: Both workflows work independently or together.
For persistent deployments, use docker-compose.yml with .env:
cp .env.example .env # Copy and edit with your passwords/API keys
docker compose up -dOr use inline environment variables for quick testing:
services:
envhaven:
image: ghcr.io/envhaven/envhaven:latest
container_name: envhaven
environment:
- PUID=1000
- PGID=1000
- TZ=Etc/UTC
- PASSWORD=password # Web UI password
- SUDO_PASSWORD=password # sudo/SSH password
volumes:
- ./config:/config # Persist home directory
ports:
- 8443:8443 # Web UI
- 2222:22 # SSH Access
restart: unless-stoppedSee Configuration Reference for all options including AI API keys and SSH setup.
SSH enables the Haven CLI and direct terminal access.
| Setup | SSH Host | Haven CLI |
|---|---|---|
| Managed | ssh-{subdomain}.envhaven.app |
haven connect . myproject-alice |
| Self-hosted | your-server:2222 |
haven connect . abc@your-server:2222 |
Self-hosted setup: Import your GitHub keys automatically:
environment:
- PUBLIC_KEY_URL=https://github.com/yourusername.keys
ports:
- "2222:22"See Configuration Reference for all SSH options.
- AI Tools Guide: Setup guides, API keys, and usage examples for all 17 AI coding tools
- Configuration Reference: Complete env vars, SSH setup, bundled tools, and compose examples
- Development Guide: Building from source, testing, and contributing
- Haven CLI: CLI source and architecture