Zero-install browser coding
Run a Linux-like sandbox in the browser — terminal, files, git, and app preview. First use downloads the WASM Linux image.
See it work
Best features, in motion
Real product captures from the desktop app, browser workspace, and cross-device flows.
Continue on any device
Start on desktop, monitor from phone, return to the same agent session — messages, stream state, and permissions stay in sync.
Agxos app creator
Describe or fork an app — the agent scaffolds it on the faux desktop and you run it immediately.
Local or cloud models
Switch OpenAI OAuth, API keys, Anthropic, Ollama, or your own compatible endpoint per task.
Agents that keep going
/loop, /goal, and /subagent drive multi-step runs instead of stopping after one reply.
Checkpoints
Save a known-good point around edits and tests — revert code and agent state when something breaks.
Workflow approvals
Declarative workflows pause at human-in-the-loop gates — you approve, then execution continues.
Manager agents
A supervisor delegates to specialized agents, coordinates plans, and recovers from stalls.
Platforms
One agent, every surface
Install the native app on Windows or Linux, use the complete browser workspace, or join Android closed testing through Google Play.
⊞Windows
›_Linux
◉Browser
△Android testing
🌐
Browser sandbox
Code without installing a daemon — terminal, git, and preview in a cached WASM Linux environment.
🎯
Real target machines
Connect a machine and let agents read files, edit code, run commands, and inspect diffs in the actual workspace.
📦
Sandboxed targets
Docker/OCI/gVisor-style isolation is in progress — marketed as preview until the release path is polished.
Preview — not a production guarantee yet.
Agent & sync
Real work, synced across devices
⚡
Multi-step agent work
Reads your repo, edits files, runs commands, and checks results — not copy-paste suggestions.
🔄
Realtime session sync
Sessions, messages, stream state, permissions, and machine status follow you across phone, web, and desktop.
↩️
Recover & continue
Encrypted session history and checkpoints let you pick up elsewhere — not live process migration when a target dies.
Security & hosting
Your server, encrypted payloads
- ✓ End-to-end encrypted sync — the Glixo server stores opaque ciphertext
- ✓ Self-host your Glixo server or stay browser-only with local providers
- ✓ No public ports required on your target machine — sync routes through your server while payloads stay encrypted
Providers
Built-in and bring-your-own
🔌
Provider catalog
Anthropic API key, OpenAI API key, OpenAI OAuth, and Ollama built in; CLIProxy and OpenAI-compatible endpoints installable.
🏠
Local + cloud
Ollama for private or cheap tasks; cloud models when you need stronger reasoning.
🧩
Provider extensions
Add adapters, manifests, secrets, and model discovery through the extension system.
Memory & agents
Teams that remember
🧠
Shared project memory
Save conventions, decisions, and environment notes — retrieve them in later sessions and workflows.
👥
Group agents
Multiple agents reason together in a shared thread over one coding task.
🎯
Magentic supervisor
A manager agent delegates to specialists, coordinates plans, and recovers from stalls.
Workflows & tools
Repeatable runs you control
📋
Declarative workflows
Steps, gates, approvals, checkpoints, and resumable execution — great for demos and production guardrails.
🛠
Tools, skills & MCP
Extension-contributed tools and MCP integrations for GitLab, Playwright, search, and more.
⌨️
Slash commands
/loop, /goal, and /subagent have full runtime handling. /schedule, /handoff, and /btw capture intent — phrase scheduling carefully until full automation ships.
🛍
Extension marketplace
Install IDE extensions and Agxos apps from the marketplace.
Two ways to work
An IDE when you want it. Agxos when you don't.
Same agent, two shells — switch whenever the job suits.
⌨️
IDE Shell
Editor, terminal, git, and agent in one place — approve the changes that matter.
🪟
Agxos
Embedded agentic OS: terminal, browser, files, editor, diff viewer, App Creator, marketplace, and native app windows.
Models
Claude and GPT, side by side
Pick a provider per session or use automatic model and reasoning selection where enabled in the UI.
Verify live auto-routing in a demo before treating it as a hard marketing claim.