_
| |_ ___ _____ _____
| _| _| _ |_| -|
|_| |_| |_____|_|___/ v0.2 — Web UI shipped
A local-first resource manager so AI coding agents can share API credentials and skill docs across every project on your machine.
One ~/.trove/ directory holds every service you use — MiniMax keys, Cloudflare tokens, Stripe rk_live, the lot. Any AI coding agent (Claude Code, Cursor, Codex, Aider) reuses the same modules across all your projects.
No inject step. No cross-agent adapters. AI is the runtime.
~/.trove/ ├── minimax/ │ ├── module.md # YAML frontmatter + skill body │ └── credentials.json # secrets (gitignored, 600 perms) ├── cloudflare/ ├── stripe/ └── ...
Reference from any project's CLAUDE.md:
@trove.md
Where trove.md lists the modules this project actually needs:
@/Users/you/.trove/minimax/module.md @/Users/you/.trove/cloudflare/module.md @/Users/you/.trove/stripe/module.md
The AI auto-loads each module, fetches credentials on demand (via jq, no env pre-export), and configures MCP servers when needed. No tooling required to start — the directory convention IS the runtime.
Run trove ui and a localhost dashboard at http://127.0.0.1:7821 lets you browse installed modules, fill credentials through a form, and install new ones from the bundled library — all without ever opening credentials.json in $EDITOR.
cd ~/.trove && trove ui # → http://127.0.0.1:7821
module.md into ~/.trove/Each module ships with a gotchas-first skill body — auth header quirks, billing pitfalls, error-code tables — so the AI doesn't have to rediscover the same trap that bit the last engineer.
trove init, no inject step, no SaaS — ever)v0.2. Format is stable. Web UI shipped with four screens: Modules grid · Module detail · Credentials form · Library. Live dogfood ongoing — see SPEC §10 for the lessons-learned log driving spec evolution.
Next: MCP-shape module template (supabase-mcp as first instance), more library modules, public OSS launch.