https://github.com/euraika-labs/pan-ui
Pan by Euraika — a self-hosted AI workspace for Hermes Agent. Chat, skills, extensions, memory, profiles, and runtime controls.
https://github.com/euraika-labs/pan-ui
agent ai-workspace euraika hermes mcp nextjs observability pan self-hosted typescript webui
Last synced: 2 months ago
JSON representation
Pan by Euraika — a self-hosted AI workspace for Hermes Agent. Chat, skills, extensions, memory, profiles, and runtime controls.
- Host: GitHub
- URL: https://github.com/euraika-labs/pan-ui
- Owner: Euraika-Labs
- License: mit
- Created: 2026-04-04T16:43:39.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2026-04-06T12:00:56.000Z (2 months ago)
- Last Synced: 2026-04-06T12:17:49.383Z (2 months ago)
- Topics: agent, ai-workspace, euraika, hermes, mcp, nextjs, observability, pan, self-hosted, typescript, webui
- Language: TypeScript
- Homepage: https://github.com/Euraika-Labs/pan-ui#readme
- Size: 3.89 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
- Codeowners: .github/CODEOWNERS
- Security: SECURITY.md
- Roadmap: docs/roadmap.md
Awesome Lists containing this project
README
# Pan by Euraika
[](https://github.com/Euraika-Labs/pan-ui/actions/workflows/ci.yml)
[](https://github.com/Euraika-Labs/pan-ui/actions/workflows/codeql.yml)
[](https://www.npmjs.com/package/@euraika-labs/pan-ui)
[](https://github.com/Euraika-Labs/pan-ui/releases)
[](./LICENSE)
A self-hosted web interface for [Hermes Agent](https://github.com/NousResearch/hermes-agent) — the AI agent by Nous Research. Chat with your agent, manage skills from the skills.sh marketplace, control extensions and MCP integrations, inspect memory, and operate profiles — all from a single dashboard with live runtime awareness.
## Why Pan
Pan is not a generic chat wrapper. It exposes the full operational surface of a running Hermes Agent instance:
- **Chat with streaming** — SSE-based streaming connected to a real Hermes runtime, with tool timelines, approval cards, and artifact rendering
- **Skills marketplace** — Browse 112+ installed skills across 27 categories, discover and install 268+ more from [skills.sh](https://skills.sh) with one click
- **MCP integrations** — View installed MCP servers, their tools, health status, and diagnostics
- **Persistent memory** — Inspect and edit both global (shared) and profile-scoped user/agent memory, with proper `§`-separated entry parsing
- **Profile isolation** — Each profile is a full workspace boundary with its own sessions, skills, memory, API keys, and policy presets
- **Runtime operations** — Approvals, run history, audit trails, telemetry, runtime health, and JSON/CSV exports
## Screenshots
### Login
### Chat
Streaming chat connected to a live Hermes runtime. Session sidebar with search, pinning, and archiving. Tool timelines expand inline. Composer shows active model, mode, tools, and profile.
### Skills — Installed
112 installed skills across 27 categories. Search by name, tag, or category. Each card shows source, tags, linked files count, and whether it's loaded in the current session.
### Skills — Discover
Browse and install skills from the skills.sh hub. Trust badges (Trusted / Official / Community), install counts, security audit results, and direct links to repos.
### Extensions & MCP
Installed MCP servers with tool inventories, health badges, and capability toggles. Test connections and view diagnostics.
### Memory
Global memory (shared across profiles) displayed as read-only cards with a blue "shared" badge. Profile-scoped memory is editable. Entries are separated by `§` and counted correctly.
### Profiles
Profile-based workspace isolation. Each profile scopes sessions, skills, memory, extensions, and API keys. Policy presets (safe-chat, research, builder, full-power) control tool access and guardrails.
### Settings
Runtime status, health monitoring, model selection, run history, audit browser, telemetry, approvals, and MCP diagnostics.
## Tech Stack
| Layer | Technology |
|-------|-----------|
| Framework | Next.js 15 (App Router) |
| Language | TypeScript |
| State | TanStack Query v5 |
| Styling | Tailwind CSS 4 |
| Testing | Vitest + Playwright |
| Runtime | Node.js 20+ |
## Quick Start
### Via npx (recommended)
```bash
npx @euraika-labs/pan-ui
```
The setup wizard will guide you through connecting to your Hermes runtime on first launch.
### From source
```bash
# 1. Clone and install
git clone https://github.com/Euraika-Labs/pan-ui.git
cd pan-ui
npm install
# 2. Start in dev mode (connects to local Hermes instance)
npm run dev
# 3. Open in browser
open http://localhost:3199
```
Default credentials: `admin` / `changeme`
### Environment Variables
| Variable | Default | Description |
|----------|---------|-------------|
| `HERMES_MOCK_MODE` | `true` | Set `false` to connect to a real Hermes instance |
| `HERMES_HOME` | `~/.hermes` | Hermes home directory |
| `HERMES_API_BASE_URL` | `http://127.0.0.1:8642` | Hermes API endpoint |
| `HERMES_API_KEY` | — | API key for Hermes (if configured) |
| `HERMES_WORKSPACE_USERNAME` | `admin` | Login username |
| `HERMES_WORKSPACE_PASSWORD` | `changeme` | Login password |
| `HERMES_WORKSPACE_SESSION_SECRET` | — | Cookie signing secret |
| `PORT` | `3199` | Pan UI port |
## Architecture
```
┌─────────────────────────────────────────────────┐
│ Browser │
│ Next.js App Router + TanStack Query + Tailwind │
└───────────────────────┬─────────────────────────┘
│ fetch / SSE
┌───────────────────────▼─────────────────────────┐
│ Next.js API Routes │
│ /api/chat/stream /api/skills /api/memory │
│ /api/profiles /api/extensions /api/runtime│
└──────┬────────────────┬─────────────────────────┘
│ │
▼ ▼
┌──────────────┐ ┌──────────────────────────────┐
│ Hermes API │ │ Hermes Filesystem │
│ :8642 │ │ ~/.hermes/ │
│ OpenAI-compat│ │ ├─ profiles/ │
│ SSE streaming│ │ ├─ skills/ │
└──────────────┘ │ ├─ memories/ │
│ └─ state.db │
└──────────────────────────────┘
```
- **Chat streaming** bridges the Hermes OpenAI-compatible `/v1/chat/completions` SSE endpoint to Pan
- **Skills** are read from `~/.hermes/skills/` with YAML frontmatter parsing; hub skills come from the skills.sh cache
- **Sessions** are read from `state.db` (SQLite) via the Hermes filesystem
- **Memory** reads/writes `USER.md` and `MEMORY.md` files at both global and profile scope
- **Profiles** map to `~/.hermes/profiles//` directories, each with their own config, memories, sessions, and skills
## Project Structure
```
src/
├── app/ # Next.js routes and API endpoints
│ ├── api/ # Server-side API routes
│ │ ├── chat/ # Chat stream, sessions
│ │ ├── skills/ # Skills CRUD, hub, categories
│ │ ├── memory/ # User/agent memory, context inspector
│ │ ├── profiles/ # Profile CRUD
│ │ ├── extensions/ # MCP extensions
│ │ └── runtime/ # Health, approvals, runs, export
│ └── [page]/ # Client page routes
├── features/ # UI feature modules
│ ├── chat/ # Chat screen, composer, transcript, tools
│ ├── skills/ # Skills browser, detail, hub cards
│ ├── memory/ # Memory editor, context inspector
│ ├── extensions/ # Extension cards, tool inventory
│ ├── profiles/ # Profile management, switcher
│ ├── sessions/ # Session sidebar, search
│ └── settings/ # Runtime, health, audit, approvals, runs
├── server/ # Server-side logic
│ └── hermes/ # Hermes filesystem bridge
├── components/ # Shared layout and UI components
├── lib/ # Types, schemas, stores, utilities
└── styles/ # Global CSS and theme variables
tests/
├── unit/ # Vitest unit tests
└── e2e/ # Playwright end-to-end tests
docs/ # Architecture, specs, and screenshots
```
## Verification
```bash
npm run lint # ESLint
npm run test # Vitest unit tests
npm run build # Production build
npm run test:e2e # Playwright e2e (requires running dev server)
```
## Security
- CLI commands use an allowlist guard (`ALLOWED_COMMANDS`) before `execFileSync` — no arbitrary command injection
- Profile isolation ensures each workspace boundary has its own sessions, memory, and API keys
- CodeQL scanning runs on every push and PR
- File path parameters are sanitized to prevent directory traversal
- Login is cookie-based with configurable credentials
## Contributing
Please read `CONTRIBUTING.md` before opening issues or pull requests.
## License
MIT