Omnara: Run Claude Code and Codex from Anywhere

Omnara is a web and mobile agentic IDE developed to allow seamless operation and interaction with Claude Code and Codex sessions from both desktop and mobile environments. It enables developers to run these sessions on their own machines and keep them accessible through a mobile and web GUI, offering flexibility when away from the desk.
The core innovation with Omnara lies in its ability to transition from local machines to a hosted remote sandbox when your device goes offline. The agent's conversation state is preserved on the server, and optional cloud syncing of working code is available. This feature utilizes git commits to maintain the state of the code base, allowing the session to continue smoothly upon shifting between local and cloud environments. Moreover, Omnara offers a voice agent for hands-free interaction, enhancing its utility while in motion.
The implementation involves a headless daemon on the user's machine, maintaining an outbound WebSocket connection with Omnara's server, facilitating uninterrupted communication between local and remote sessions. This setup obviates the need for exposed ports or SSH access, enhancing security and simplicity. Users can begin by executing the following command to install Omnara:
curl -fsSL https://omnara.com/install/install.sh | bashAfter installation, running omnara inside any git repository starts a headless Claude Code or Codex session linked to the Omnara web and mobile apps. The platform is available for free for up to 10 agent sessions monthly, with a $20/month subscription for unlimited sessions.
📖 Read the full source: HN AI Agents
👀 See Also

WinRemote MCP: Open Source MCP Server for Full Control of Windows Desktops
WinRemote MCP provides AI agents with full control over Windows desktops, allowing for UI detection, file operations, registry access, and more, utilizing over 40 tools.

Claude-Real-Video: Scene-Aware Frame Extraction + Transcript for Any LLM
An open-source tool that extracts scene-aware, deduplicated frames and audio transcripts from videos, enabling any LLM to "watch" a video locally without uploading. Features scene-change detection, sliding-window dedup, and Whisper transcription.

Benchmark Results: 331 GGUF Models Tested on Mac Mini M4 16GB
A benchmark of 331 GGUF models on a Mac Mini M4 with 16GB RAM reveals only 11 Pareto-optimal models, all Mixture-of-Experts architectures. Mixture-of-Experts models dominate performance with median 20.0 tokens/second versus 4.4 for dense models.

Colony: A Local-First Coordination Layer That Cuts Multi-Agent Handoff Tokens from 30K to 400
Colony is a local-first coordination substrate that reduces multi-agent handoff costs from ~30,000 tokens to ~400 by replacing context replay with compact observations stored in SQLite.