Local models · every device

GlixoCode

A coding agent that stays with your work across desktop, browser, and Android.

Download Glixo Code Checking latest release ▶ Try in browser
WindowsLinuxBrowserAndroid
Glixo Code desktop — a live coding-agent session with changed files and commits
See it work

Real sessions, not mockups.

One session, any provider.

Start on Gemini, switch to Claude or GPT mid-task — the same session, history, and tools carry over. Or let auto-routing pick a provider per message.

Point it at a goal and let it loop.

/loop ×3 re-runs the task for a bounded number of rounds — here it reviews a reconstruction pipeline three times, reporting one concrete finding per pass, then stops.

Set the outcome, not every step.

Give it a goal — “harden the webhook retry path with a focused regression test” — and it edits, runs the tests, and keeps going until they pass.

Everything the agent needs, in one place.

Real product surfaces — a desktop app, a full browser workspace, and cross-device sync — not copy-paste suggestions.

Browser Linux workspace in Glixo Code

Zero-install browser coding

A full Linux workspace in a browser tab — terminal, git, and live preview, in a cached WASM image.

Provider and model picker with auto-routing

Local or cloud models

Pick a provider or let it auto-route — Anthropic, OpenAI, Gemini, Ollama, and your own endpoints.

Declarative workflow with a human approval gate

Workflows & approvals

Declarative workflows pause at human-in-the-loop gates — you approve, then execution continues.

Glixo Code settings — providers, memory, MCP, extensions

Extend it — providers, memory & MCP

Accounts, session defaults, memory, MCP servers, and installable extensions, all in one place.

Works with your models
Specialized sandbox image

Pre-baked for speed — and fewer tokens.

Toolchains, caches, and CLIs are baked into the sandbox image, so the agent runs real work instead of reinstalling it — and reaches for compact commands instead of verbose host shells.

Glixo Code Machines dashboard: fleet status with machines, sandboxes, and direct hosts, plus per-machine hosts and sandbox environments across a Windows workstation and a Linux render node
2.30×
as fast
1.91s in the sandbox vs 4.41s in PowerShell, across a six-task workload.
72.5%
fewer command tokens
106 vs 385 command-text tokens — 279 saved every run.
How it was measured
  • Measured on Windows via real Glixo RPC across a six-task workload.
  • Windows PowerShell: 4,406.5 ms · Sandbox: 1,913.5 ms (2,493 ms faster) — about 2.30× the workloads in the same time.
  • Five of six tasks were faster (jq ran 6.62× faster); cloc was 63.1% slower.
  • Command-text tokens: 385 → 106, saving 279 (72.5%).

Your server. Encrypted payloads.