Repository navigation
Releases: DayZAnder/llm-os
Release list
v0.4.4
Full Changelog: v0.4.3...v0.4.4
v0.4.3
v0.4.2
v0.4.1
Warning
The Server and Desktop images in this release don't boot: they stop at grub rescue>. None of the images in this release power off when the hypervisor sends a shutdown, so settings saved in the last ~30 seconds can be lost. Use v0.4.2 or later; your llmos-data disk works with it unchanged.
Full Changelog: v0.4.0...v0.4.1
v0.4.0
v0.3.0
LLM OS v0.2.4 — Prompt Router, KasmVNC Browser, SSH Persistence
What's New
Intelligent Prompt Router
A small LLM (0.5B) classifies every prompt — type (iframe/process), complexity, template match, and model hint. Falls back to regex when no LLM is available. Say "using opus" or "with haiku" to pick a specific model.
Resource Monitor
Probes available models every 60 seconds, ranks by tier (1-9), and auto-selects the strongest model for each task. Add a GPU with Ollama → the OS gets smarter. Add a Claude API key → smarter again. Remove them → graceful degradation.
KasmVNC Browser
Replaced the 4-process noVNC stack (Xvfb + x11vnc + websockify + noVNC) with KasmVNC — a single binary that handles X server, VNC encoding, WebSocket transport, and web client. Faster, smoother, 60fps.
SSH Client Persistence
SSH sessions survive page refreshes. The server maintains a session map with scrollback buffers. Reconnecting replays terminal history and resumes the live session. Sessions auto-clean after 10 minutes of inactivity.
Known App Templates
SSH client, Chromium browser, and NanoClaw now skip LLM generation entirely — they launch pre-defined Docker blueprints instantly.
GitHub Pages
- Screenshots gallery with lightbox viewer
- Vision & realistic targets timeline
- Best use cases section
Other
- 355 tests passing, zero runtime dependencies
- Private info removed from build scripts
- Updated CONTRIBUTING.md with current help-wanted priorities
VM Images
Same VM images as v0.2.3 — the kernel, Docker, and Node.js haven't changed. Only the LLM OS application code is updated (auto-pulled on boot if network is available).
| Variant | Use Case |
|---|---|
| Server (~700MB) | Full features, headless, Docker + SSH |
| Desktop (~1GB) | Server + Chromium kiosk boot |
| Micro (~50MB) | Minimal, iframe apps only, no Docker |
Formats: QCOW2 (Proxmox/KVM), VHDX (Hyper-V), OVA (VirtualBox)
Default login: root / llmos — change on first boot!
Full Changelog: v0.2.3...v0.2.4
v0.2.3
What's New in v0.2.3
Phase 3a: Security & Sandboxing
- WASM Sandbox: Server-side WebAssembly sandbox using
worker_threads+SharedArrayBufferIPC for isolated compute - Cryptographic Capability Tokens: HMAC-SHA256 signed tokens for fine-grained app permissions (
POST /api/grant)
New Features
- App Inject API:
POST /api/injectendpoint for externally-generated apps (inject from CLI or scripts) - GitHub Actions CI: Automated VM image builds (micro, server, desktop) on push
Fixes
- Fix shell crash on clarification responses (
needsClarificationJSON parse error) - Fix Buildroot boot: Node 20 LTS (Node 22 breaks with musl), syslinux config, dropbear symlinks
Docs & Infra
- Updated CLAUDE.md, README, CONTRIBUTING, and landing page
- QC prompts added to CONTRIBUTING.md
- Build output directory added to .gitignore
VM Images: micro (Buildroot, ~50MB), server (Alpine + Docker), desktop (Alpine + Docker + Cage/Chromium kiosk)
Full Changelog: v0.2.2...v0.2.3
LLM OS v0.2.1 — Multi-Provider Gateway & Self-Improvement
What's New
- Multi-provider LLM gateway: Ollama, Claude, OpenAI — auto-routing by complexity with fallback
- Claude Code agent: Self-improvement scheduler can spawn Claude Code for multi-file fixes
- GPL compliance: THIRD-PARTY-LICENSES.md for Buildroot/busybox/Linux
- Buildroot micro build fix: Correct config file paths
- Phase 3 roadmap update: Split into 3a (WASM + system config) and 3b (unikernel)
Full Changelog
LLM OS v0.2.0 — Multi-Format Build
What's New
- Multi-format VM builds: QCOW2 (Proxmox/KVM), VHDX (Hyper-V), OVA (VirtualBox)
- Version bump across all build scripts and boot screens