geeksy

Chat interface for autonomous AI agents. Built with Melina.js, powered by smart-agent and jsx-ai.

┌────────────────────────────────────────────────────────────────┐
│  Geeksy                                          gemini-2.5 ▼ │
├──────────┬─────────────────────────────────────────────────────┤
│ Agents   │                                                    │
│          │                  🤖 Geeksy                         │
│ + New    │     Create an agent and describe what you want.     │
│          │                                                    │
│          │   [tell me a joke]  [list files]  [create file]    │
│          │                                                    │
│          ├─────────────────────────────────────────────────────┤
│          │  Type your command or question...            [➤]   │
│          ├─────────────────────────────────────────────────────┤
│ Models   │  Objectives │ Files │ Schedule                     │
│ Skills   │  No objectives yet.                                │
│ Plugins  │                                                    │
├──────────┴─────────────────────────────────────────────────────┤
│ ⚙ Settings                                                    │
└────────────────────────────────────────────────────────────────┘

Quick Start

bun install
bun run dev
# → http://localhost:3737

Set an API key in the Settings panel or via environment:

export GEMINI_API_KEY=your-key
bun run dev

How It Works

1. Create an agent — click "New Agent", pick a model, add skills
2. Chat — type a message, the agent plans objectives and executes them
3. Watch — objectives, tool calls, and file changes stream in real-time via SSE

Under the hood:

• smart-agent runs the agentic loop — planner generates objectives, worker executes tools, validator checks success
• jsx-ai handles LLM calls — provider routing (Gemini, OpenAI, Anthropic, DeepSeek), streaming, retries
• Melina.js serves the frontend — SSR + vanilla client runtime, zero React on the client

Project Structure

app/                    # Melina.js chat frontend
├── server.ts           # Entry point — starts Melina on port 3737
├── page.tsx            # Main chat page (SSR)
├── globals.css         # Dark theme styles
├── lib/                # Client-side modules
│   ├── agents.tsx      # Agent CRUD + chat logic
│   ├── chat.tsx        # SSE streaming + message rendering
│   └── settings.tsx    # API key management
└── api/                # Backend API routes
    ├── chat/route.ts   # SSE streaming chat endpoint (Session pipeline)
    ├── agents/route.ts # Agent management (in-memory)
    ├── models/route.ts # Model listing + API key storage
    ├── skills/route.ts # Skill file discovery
    └── schedule/route.ts

API

| Endpoint | Method | Description | |----------|--------|-------------| | /api/chat | POST | Start a chat session, returns SSE stream | | /api/agents | GET/POST/PUT/DELETE | Agent CRUD | | /api/models | GET | List available LLM models | | /api/models/keys | GET/POST | API key management | | /api/skills | GET | List available skill files |

Dependencies

| Package | Role | |---------|------| | @mements/smart-agent | Agentic loop — Agent, Session, objectives, skills | | jsx-ai | LLM primitives — callLLM, callText, streamLLM | | melina | Web framework — SSR, routing, client runtime | | measure-fn | Performance instrumentation | | satidb | SQLite persistence (Zod schemas) |

License

MIT