open-hub

Open-Hub

---

Open-Hub is a terminal AI agent that writes code, edits files, runs commands, controls your OS, and autonomously completes complex multi-step tasks — all from a single ohagent command.

No cloud dashboard. No subscriptions. Bring your own API key.

Features

• 17 AI providers out of the box — Anthropic, OpenAI, Google Gemini, Groq, Mistral, xAI, Perplexity, DeepSeek, Together AI, Fireworks, Cerebras, Cohere, Novita, Qwen, Ollama, Azure OpenAI, plus any custom OpenAI-compatible endpoint
• Interactive model & key manager — switch models and set API keys directly from the TUI with /model and /keys
• MCP support — connect any Model Context Protocol server with one command or env var
• Goal-driven mode — set a goal, the agent loops autonomously until it's done
• 24 built-in tools — file system, shell, OS-level (clipboard, screenshots, notifications, AppleScript), process management
• Session management — save, resume, and name conversations
• Context compaction — automatically summarize long conversations to stay within model context limits
• Dynamic providers — add any OpenAI-compatible API via env vars, no code changes needed
• Full system access — read/write any file, run any command, control any app

Quick Start

# Install globally
npm install -g @sultonsho-nazarshoev/open-hub

# Set an API key (pick any provider)
export ANTHROPIC_API_KEY=sk-ant-...

# Launch
ohagent

That's it. No config file needed on first run.

Installation

Prerequisites

• Node.js 18+
• npm 9+

From npm (recommended)

npm install -g @sultonsho-nazarshoev/open-hub

From source

git clone https://github.com/dark-developer-lord/open-hub.git
cd open-hub
npm install
npm run build   # TypeScript → dist/
npm link        # registers `ohagent` globally

Configuration

On first run, Open-Hub creates ~/.open-hub/config.json automatically.

API keys via environment variables

export ANTHROPIC_API_KEY=sk-ant-...
export OPENAI_API_KEY=sk-...
export GOOGLE_API_KEY=...
export GROQ_API_KEY=...

API keys via TUI

You can add or update API keys without leaving the terminal — just type /keys inside a session:

/keys          → opens provider list; select a provider and paste your key

Keys are saved to ~/.open-hub/config.json and take effect immediately.

Other configuration variables

# Defaults
DEFAULT_PROVIDER=anthropic
DEFAULT_MODEL=claude-opus-4-5

# Limits
MAX_TOKENS_PER_TURN=32768       # output tokens per model call
MAX_GOAL_ITERATIONS=200         # max iterations in --goal mode

# Permission mode
AGENT_PERMISSION_MODE=bypassPermissions

Usage

Interactive TUI

ohagent

Starts a full-screen TUI session. Type your task and the agent executes it using tools.

One-shot mode

ohagent -p "explain this codebase"
ohagent -p "fix all TypeScript errors in src/"

Goal-driven autonomous mode

The agent loops until the goal condition is satisfied:

ohagent --goal "all tests pass and there are no TypeScript errors"
ohagent --goal "implement the feature described in TASK.md" --max-iterations 300

Specify model and provider

ohagent -m gpt-4o --provider openai
ohagent -m claude-opus-4-5
ohagent -m gemini-2.5-pro --provider google
ohagent -m llama-3.3-70b-versatile --provider groq

Session management

ohagent -c                  # continue last session
ohagent -r my-project       # resume session by name or ID

All CLI flags

Flag	Description
-p, --print <query>	Non-interactive: run query and exit
-m, --model <model>	Model to use
--provider <name>	Provider to use
-c, --continue	Continue most recent session
-r, --resume <id>	Resume session by ID or name
--goal <condition>	Goal-driven autonomous mode
--max-tokens <n>	Max output tokens per turn (default: 32768)
--max-iterations <n>	Max goal loop iterations (default: 200)
--permission-mode <mode>	bypassPermissions | default | plan
--mcp-config <path>	Load MCP servers from JSON file
--verbose	Show detailed tool execution output

TUI commands

Type these inside an interactive session:

Command	Description
/help	Show all commands
/model	Open interactive model selector (↑↓ to navigate, Enter to confirm)
/model <name>	Switch model directly by name
/keys	Open API key manager (add or update keys per provider)
/provider <name>	Switch provider
/providers	List all available providers
/goal <condition>	Start autonomous goal loop
/mcp list	List connected MCP servers
/mcp install <name>	Install a popular MCP server
/mcp catalog	Browse popular MCP servers
/provider add	Add a custom provider interactively
/integrations	Show all active providers and MCP servers
/sessions	List saved sessions
/save [name]	Save current session
/compact	Summarize conversation to free context
/tokens	Show token usage
/exit	Exit

Providers

Provider	Models	Notes
Anthropic	Claude Opus 4.5, Sonnet 4.5, Haiku 3.5	Best for complex tasks
OpenAI	GPT-4o, o3, o4-mini, GPT-4.1
Google	Gemini 2.5 Pro/Flash, 2.0 Flash	Largest context
Groq	Llama 4, Llama 3.3, Qwen 2.5	Fastest inference
Mistral	Magistral, Mistral Large/Medium/Small
xAI	Grok-3, Grok-3-mini
Perplexity	Sonar Large/Small	Web-search enabled
DeepSeek	DeepSeek R2, V3	Cost-effective
Together AI	500+ open-source models
Fireworks	Llama, Qwen, DeepSeek
Cerebras	Llama 3.3 70B	Hardware-accelerated
Cohere	Command A+, Command R+
Novita AI	Open-source models
Qwen	Qwen3, QwQ, Qwen2.5-Coder
Ollama	Any local model	No API key needed
Azure OpenAI	GPT-4o, o-series	Enterprise
Custom	Any OpenAI-compatible API	Via env vars

Add a custom OpenAI-compatible provider

Works with LM Studio, vLLM, OpenRouter, and any other OpenAI-compatible endpoint:

PROVIDER_OPENROUTER_BASE_URL=https://openrouter.ai/api/v1
PROVIDER_OPENROUTER_API_KEY=sk-or-...
PROVIDER_OPENROUTER_NAME=OpenRouter
PROVIDER_OPENROUTER_MODELS=meta-llama/llama-4-scout,google/gemini-2.5-flash

MCP (Model Context Protocol)

Connect external tools and data sources:

# Inside a session:
/mcp install memory          # persistent memory across sessions
/mcp install filesystem      # extended file access
/mcp install github          # GitHub integration
/mcp catalog                 # browse 13 popular servers

Or configure via ~/.open-hub/mcp.json or env vars:

MCP_SERVER_MEMORY_TYPE=stdio
MCP_SERVER_MEMORY_COMMAND=npx
MCP_SERVER_MEMORY_ARGS=-y,@modelcontextprotocol/server-memory

Built-in Tools

Shell

• Bash — run any command with full shell features

File System

• Read — read any file with optional line ranges
• Write — create or overwrite files (auto-creates directories)
• Edit — precise string replacement in files
• Delete — delete files or directories
• Move — move or rename
• Copy — copy files or directories
• MakeDir — create directory trees
• FileInfo — metadata (size, permissions, timestamps)
• Chmod — change permissions
• LS — list directory contents
• Glob — find files by pattern
• Grep — search file contents

System / OS (macOS + Linux)

• Open — open files, URLs, or applications
• AppleScript — automate any macOS app (Finder, Safari, Mail, etc.)
• ClipboardRead / ClipboardWrite — system clipboard
• Notify — desktop notifications
• Screenshot — capture screen or windows
• ProcessList — list running processes
• KillProcess — send signals to processes
• SystemInfo — CPU, memory, disk, network info
• ListApps — installed applications
• Env — read environment variables

Development

git clone https://github.com/dark-developer-lord/open-hub.git
cd open-hub
npm install

# Dev mode (no build step)
npm run dev -- -p "hello"

# Type-check
npm run lint

# Build
npm run build

Project structure

src/
├── index.ts              # CLI entry point & provider wiring
├── agent/
│   ├── Agent.ts          # Core agent loop
│   └── GoalManager.ts    # Goal-driven execution
├── providers/            # One file per AI provider
│   ├── OpenAICompatProvider.ts  # Abstract base class
│   └── ...
├── tools/
│   ├── BashTool.ts
│   ├── FileTool.ts       # All file system tools
│   ├── SystemTool.ts     # All OS-level tools
│   └── ToolRegistry.ts
├── mcp/
│   └── MCPManager.ts     # MCP client manager
├── commands/
│   └── CommandRegistry.ts # Slash commands
├── config/
│   └── Config.ts         # ~/.open-hub/config.json
├── session/
│   └── SessionManager.ts
└── ui/
    └── TUI.tsx           # ink-based terminal UI

Contributing

Pull requests are welcome. For major changes, open an issue first.

License

MIT