opencode-workspace

My personal AI workspace: an OpenCode-centered runtime with custom skills, long-term memory, notes, scheduling, and email/Telegram bridges. It can also route work through Claude when needed.

What it does

• Runs a project-local OpenCode server and stores sessions in this workspace
• Supports 2 runtimes: OpenCode and Claude
• Provides a built-in Gmail bridge
• Keeps a set of skills customized to my needs
• A Cloudflare Worker endpoint that exposes the agent's live status and activity — see hcxu.cc/agent
• Assumes a git repo checked out into the notes/ folder at the root of this repo

How it fits together

• This repo — the workspace runtime: code, skills, instructions, bridge services, and project-local OpenCode state
• notes/ — a separate private git repo checked out here, holding my personal data, memory, research, and todos. Kept separate so this repo stays code, and notes/ stays the data it operates on.
• Bridges — the Gmail bridge and opencode-telegram-bot let me reach the workspace when I'm away from the desk
• Status worker — a Cloudflare Worker publishing live status to hcxu.cc/agent

Quick start

npm install
cp .env.example .env         # then fill it in (see below)
npm run bridge               # start the workspace bridge and OpenCode server
npm start                    # run only the OpenCode server
npm run tui                  # open the OpenCode TUI against the project-local DB

Key .env settings (all documented inline in .env.example):

• Gmail bridge — AGENT_INBOX_EMAIL, USER_EMAIL, and an app password (EMAIL_PASSWORD), not your login password
• Notes repo — NOTES_REPO_URL (+ NOTES_REPO_TOKEN for HTTPS auth); then run scripts/bootstrap-notes.sh to check it out into notes/
• Long-term memory — GOOGLE_API_KEY (Gemini, for embeddings + extraction) and a running Qdrant at QDRANT_URL

There's no first-run setup wizard yet — this is manual. For VPS deployment, see docs/DEPLOY.md and the deploy-agent skill.

How I use it

My main entry point is the opencode-telegram-bot, which talks to the OpenCode server running from this workspace. This lets me keep working from Telegram when I'm away from the desk.

When I'm at the machine, I run npm run tui in this repo. The TUI uses the same project-local OpenCode state as the Telegram bot, so both sides see the same sessions and I can continue the same work from either Telegram or the local terminal.