A self-hosted prompt journal + CLI that captures every interaction
with Claude Code, Codex, OpenCode, Gemini CLI, and more — then turns them into actionable insights.
Quickstart for Agents · Start with Human · CLI · Dashboard · Local Mode · Server Deploy · Realtime: muxa · Contributing
You write hundreds of prompts a day to AI coding agents. But do you actually know which ones work?
Oh My Prompt gives you the answer. It captures every prompt, scores its quality, and shows you patterns you'd never notice on your own.
|
🎯 Capture Shell hooks silently intercept |
📊 Analyze Quality scores, token usage, |
🔄 Sync Local SQLite → server API |
| Claude Code Anthropic |
Codex OpenAI |
OpenCode Open Source |
Gemini CLI |
| Hook + Backfill | Hook + Backfill | Hook + Backfill | Hook + Backfill |
All agents are auto-detected during
omp setup. Hooks capture prompts in real-time; backfill imports your full history.
Install and set up Oh My Prompt (https://github.com/jiunbae/oh-my-prompt).
npm install -g oh-my-prompt && omp setup
Run `omp setup` interactively — do NOT pass any flags.
Ask me for each input (server URL, auth, device name, hooks).
When done, run `omp doctor` and summarize the result.
Setup Flow Preview
omp setup launches an interactive wizard with colored prompts, spinners, and a summary box:
$ omp setup
┌ oh-my-prompt
│
◆ Server URL
│ https://prompt.jiun.dev
│
◆ Authentication
│ ● Login with email & password (recommended)
│ ○ Paste existing API token
│
◆ Email
│ user@example.com
│
◆ Password
│ ********
│
◇ Authenticating... done
│
◆ Device name
│ my-macbook
│
◇ Database migrated (schema v3)
│
◆ Install hooks
│ ◻ Claude Code (detected)
│ ◻ Codex (detected)
│ ◻ OpenCode (not found)
│ ◻ Gemini CLI (detected)
│
◇ Hooks installed (Claude Code, Codex, Gemini CLI)
│
◇ Server validated (200)
│
◇ Setup Complete ───────────────────╮
│ │
│ Server: https://prompt.jiun.dev │
│ Device: my-macbook │
│ Hooks: claude, codex, gemini │
│ │
├────────────────────────────────────╯
│
└ Run omp backfill to import existing prompts
Non-interactive mode (--yes) and JSON output (--json) are fully supported for CI/scripting.
Manual Install Options
# npm (recommended)
npm install -g oh-my-prompt && omp setup
# npx (no global install)
npx oh-my-prompt setup
# source
git clone https://github.com/jiunbae/oh-my-prompt.git
cd oh-my-prompt
pnpm install
pnpm build:cli
cd packages/omp-cli
npm link
omp setup You CLI Dashboard
─── ─── ─────────
claude "fix the bug"
│
└──── hook ────▶ omp ingest ──▶ SQLite (local)
│
├── omp sync ──▶ POST /api/sync/upload
│ │
│ ┌──────┴──────┐
│ │ PostgreSQL │
│ └─────────────┘
│ │
│ ┌────────┴────────┐
│ │ omp serve │ ← local mode
│ │ localhost:3000 │
│ └─────────────────┘
│ or
│ ┌─────────────────┐
└──────────────│ your-server.com │ ← server mode
└─────────────────┘
# Install
npm install -g oh-my-prompt
# Setup (interactive wizard)
omp setup
# Verify
omp doctorThat's it. Now use Claude Code, Codex, OpenCode, or Gemini CLI normally — prompts are captured automatically.
claude "Refactor this function to use async/await"
# ↑ captured silently in the background| SQLite (no Docker) | Local Docker | Server | |
|---|---|---|---|
| Setup | omp serve --local |
omp serve |
Deploy to your server |
| Requires | Nothing | Docker | Docker + domain |
| Dashboard | localhost:3000 |
localhost:3000 |
your-domain.com |
| Database | SQLite (same as CLI) | PostgreSQL + Redis | PostgreSQL + Redis |
| Data | Local only | Local only | Multi-device sync |
| Best for | Solo use, minimal setup | Full features, solo | Teams, cross-machine |
SQLite Mode — zero dependencies, view your data instantly:
omp serve --local # Dashboard at http://localhost:3000
omp serve --local --port 8080 # Custom port
omp serve --local --host 0.0.0.0 # Network accessDocker Mode — full-featured with PostgreSQL and Redis:
omp serve # Start dashboard at http://localhost:3000
omp sync # Sync captured prompts to local dashboardServer Mode — deploy once, sync from anywhere:
omp config set server.url https://your-domain.com
printf '%s' "$OMP_TOKEN" | omp config set server.token --stdin
omp sync # Sync to remote serveromp setup — Interactive configuration wizard
Beautiful step-by-step wizard powered by @clack/prompts:
$ omp setup
┌ oh-my-prompt
│
◆ Server URL .............. https://your-server.com
◆ Authentication .......... Login with email & password
◆ Email ................... you@example.com
◆ Password ................ ********
◇ Authenticating... Logged in as you@example.com
◆ Device name ............. my-laptop
◇ Database migrated (schema v3)
◆ Install hooks ........... Claude Code, Codex, Gemini CLI
◇ Hooks installed (Claude Code, Codex, Gemini CLI)
◇ Server validated (200)
│
◇ Setup Complete ───────────────────╮
│ │
│ Server: https://your-server.com │
│ Device: my-laptop │
│ Hooks: claude, codex, gemini │
│ │
├────────────────────────────────────╯
│
└ Run omp backfill to import existing prompts
Flags: --yes (non-interactive), --json (machine-readable), --skip-validate, --server, --token, --device
omp analyze — Score prompt quality
$ omp analyze abc123
Score: 85/100 (Good)
Signals:
✓ Goal Clear objective stated
✓ Context Background information provided
✗ Constraints No specific constraints
✓ Output Expected format described
✗ Examples No examples included
Suggestions:
→ Add specific constraints or requirements
→ Include examples of expected outputomp stats — View statistics
$ omp stats --view weekday --since 7d
┌──────────────────────────────────────────────────────────────┐
│ Local Analytics │
│ Range: 7d -> now │
└──────────────────────────────────────────────────────────────┘
Top Projects
api ██████████████████████████ 42 6.2k tok · 88% rsp
frontend ████████████████░░░░░░░░░ 27 3.8k tok · 91% rsp
Grouped By weekday
Mon ██████████████████████████ 18 2.1k tok · 94% rsp
Tue ████████████████████░░░░░░ 14 1.8k tok · 86% rspAll commands
| Command | Description |
|---|---|
omp setup |
Interactive configuration wizard |
omp install [claude|codex|opencode|gemini|all] |
Install capture hooks |
omp uninstall [claude|codex|opencode|gemini|all] |
Remove capture hooks |
omp status |
Show config and hook status |
omp doctor |
Validate setup, diagnose issues |
omp sync |
Sync local prompts to server |
omp sync status |
Show sync history |
omp backfill [--claude-only|--codex-only|--opencode-only|--gemini-only] |
Import history from all agents |
omp serve |
Start local dashboard (Docker) |
omp serve --local |
Start local dashboard (SQLite, no Docker) |
omp serve stop |
Stop local dashboard server |
omp stats [--view overview|projects|sources|hourly|weekday|sessions] [--group-by day|week|month|project|source|hour|weekday] |
View local analytics |
omp export [--format json|jsonl|csv] |
Export prompts |
omp import codex-history |
Import from Codex |
omp config get|set|validate |
Manage configuration |
omp db migrate |
Run database migrations |
The self-hosted web dashboard turns raw prompts into insights.
|
Prompt Journal
|
Analytics
|
|
Multi-User
|
Security
|
Admin Diagnostics — admins can open /admin/diagnostics to check database,
Redis, LLM configuration, optional live LLM probe, timezone/KST handling, i18n
locale support, sync recency, and AI insight cache health. The diagnostics API
only returns redacted configuration state; secrets and API keys are never shown.
Screenshots
| Analytics | Sessions |
| AI Insights | Search |
| Templates | Login |
Light Mode
| Dashboard | Analytics |
The fastest way to view your data. No Docker, no server — just your existing SQLite database.
omp serve --local # Dashboard at http://localhost:3000
omp serve --local --port 8080 # Custom port
omp serve --local --host 0.0.0.0 # Expose to networkThis reads directly from the same SQLite database the CLI uses. No sync, no registration — just open and browse.
For the full dashboard experience with PostgreSQL, Redis, and the web UI:
# Start (pulls images and runs PostgreSQL + Redis + App)
omp serve
# Dashboard is now at http://localhost:3000
# Register an account, then:
omp config set server.url http://localhost:3000
printf '%s' "$OMP_TOKEN" | omp config set server.token --stdin # from Settings page
omp backfill # import history from all agents
omp sync # sync to local dashboardomp serve status # check container status
omp serve logs # tail app logs
omp serve stop # stop (data is preserved)
omp serve # restart — your data is still thereConfiguration:
omp config set serve.port 3030 # change port (default: 3000)
omp config set serve.adminEmail you@email.com # auto-seed admin account
omp config set serve.image my-registry/omp # custom Docker imageFor multi-device sync and team use, deploy Oh My Prompt to your own server.
git clone https://github.com/jiunbae/oh-my-prompt.git
cd oh-my-prompt
docker compose up -d # Starts PostgreSQL + Redis + App on :3000docker run -p 3000:3000 \
-e DATABASE_URL=postgresql://user:pass@host:5432/prompts \
-e SESSION_SECRET=$(openssl rand -hex 32) \
-e OMP_ADMIN_EMAIL=you@email.com \
registry.jiun.dev/oh-my-prompt:latest| Variable | Required | Default | Description |
|---|---|---|---|
DATABASE_URL |
Yes | — | PostgreSQL connection string |
SESSION_SECRET |
Yes | — | Cookie signing key (openssl rand -hex 32) |
REDIS_URL |
No | redis://localhost:6379 |
Redis for caching |
OMP_ADMIN_EMAIL |
No | — | Auto-seed admin email on startup |
NODE_ENV |
No | production |
Environment mode |
NEXT_PUBLIC_APP_URL |
No | — | Public base URL for absolute links |
COOKIE_SECURE |
No | true in production |
Set false only for plain-HTTP deployments |
OMP_TIME_ZONE |
No | TZ or UTC |
App analytics timezone, e.g. Asia/Seoul |
AI insights use an OpenAI-compatible or provider-specific LLM configured by environment variables:
| Variable | Required | Default | Description |
|---|---|---|---|
OMP_LLM_PROVIDER |
No | — | openai, custom, ollama, anthropic, azure, or gemini |
OMP_LLM_API_KEY |
Provider-dependent | — | API key; optional for local custom/ollama servers that do not require auth |
OMP_LLM_BASE_URL |
No | provider default | Base URL for OpenAI-compatible/local serving |
OMP_LLM_MODEL |
No | provider default | Model name |
OMP_LLM_MAX_TOKENS |
No | 2048 |
Max output tokens for insight generation |
OMP_LLM_TEMPERATURE |
No | 0.3 |
Sampling temperature |
OMP_LLM_ENABLE_THINKING |
No | — | Optional true/false flag for compatible local models |
For a local OpenAI-compatible server:
OMP_LLM_PROVIDER=custom
OMP_LLM_BASE_URL=http://localhost:8000/v1
OMP_LLM_MODEL=your-model
# OMP_LLM_API_KEY can be omitted if your local server accepts unauthenticated requests.Use /admin/diagnostics as an admin to verify the active LLM config. Click
Run LLM Probe only when you want to send a short live test request to the
configured model.
Example manifests are in k8s/. Update secrets and ingress, then select an
immutable image tag before applying the Kustomize bundle:
cd k8s
kustomize edit set image registry.jiun.dev/oh-my-prompt=registry.jiun.dev/oh-my-prompt:<git-sha>
kubectl apply -k .omp config set server.url https://your-domain.com
printf '%s' "$OMP_TOKEN" | omp config set server.token --stdin # after registration
omp backfill # import history from Claude, Codex, OpenCode, Gemini
omp sync # upload to serveroh-my-prompt/
├── src/app/ Next.js 16 App Router
│ ├── (dashboard)/ Protected pages (prompts, analytics, admin)
│ └── api/ REST API (auth, sync, analytics)
├── src/components/ React + Recharts + Shadcn/ui
├── src/db/ Drizzle ORM schema (PostgreSQL)
├── src/services/ Business logic (upload, sync, classify)
├── src/omp/ CLI source (Node.js + SQLite)
├── packages/omp-cli/ Standalone npm package
└── .gitea/workflows/ Gitea CI/CD (Docker build + k8s deploy)
| Frontend | Next.js 16 · React 19 · Tailwind CSS 4 · Recharts |
| Backend | Next.js API Routes · tRPC · Zod |
| Database | PostgreSQL · Drizzle ORM |
| CLI | Node.js · SQLite WAL (sql.js fallback) · @clack/prompts · picocolors |
| Infra | Docker · Kubernetes · ArgoCD · Gitea CI |
muxa is omp's natural counterpart on the realtime side. Where omp captures every prompt over time and turns the history into searchable analytics, muxa watches your tmux panes live — agent state (working / waiting / idle / error), pane-to-session correlation, status-line glyphs, a fullscreen TUI, and an HTTP dashboard that surfaces every tmux pane on your machine in one tab.
Same four agents on both sides; opposite axes — omp owns the history layer, muxa owns the live layer. They're complementary, not competitive.
muxa can also push PromptSubmitted events straight into your omp instance, so your prompt journal stays current without an extra hook setup. Add this to muxa's config:
[sinks.oh_my_prompt]
enabled = true
endpoint = "https://prompt.jiun.dev" # or your self-hosted URL
# token comes from $OMP_SERVER_TOKENThen run muxad and prompts flow into omp's ingest endpoint with retry, backoff, and dedup. See muxa's docs/SINKS.md for the full setup.
git clone https://github.com/jiunbae/oh-my-prompt.git
cd oh-my-prompt
pnpm install
pnpm dev # Web dashboard
pnpm build:cli # Build CLI package- Fork → 2. Branch (
feat/thing) → 3. Commit → 4. PR