Forge Remote: putting Claude Code in your pocket

Claude Code is the most powerful AI-assisted development tool I use — but it lives in a desktop terminal. Generic mobile ChatGPT apps don't solve this. They don't know your project structure, can't read your files, can't run commands, can't maintain session context across a working session.

The need: something that speaks Claude Code's language, respects project context, lets me manage development workflows from mobile with proper security — not a consumer chat wrapper with no awareness of what I'm building.

Mobile app built on Anthropic SDK directly — not a third-party wrapper, not a proxy, native SDK integration. Architecture: mobile client connects through WireGuard VPN to a self-hosted backend, which orchestrates Claude sessions using the Anthropic SDK. The codebase never leaves infrastructure I control.

Voice pipeline: Whisper for transcription, Claude for reasoning and code generation, ElevenLabs for voice response. Speak a request, hear the answer. Works on commute. Works in a meeting break. Works anywhere I don't want to open a laptop.

Security architecture: WireGuard VPN endpoint I control, client codebases never on shared servers, no third-party service except Anthropic and my own VPN endpoint. Built for someone who handles other people's codebases professionally.

Built without a formal roadmap — started as a weekend exploration and became a real tool. That was fine for getting started, but it meant some features were built in the wrong order. Should have spent a day on feature spec with explicit prioritization before writing the first component.

Should have invested earlier in automated testing. Mobile + voice + VPN + Anthropic SDK is a lot of failure surfaces in one stack. Manual testing caught most issues but added uneven QA time at every feature addition. End-to-end tests would have caught regressions faster.