Claudex

Name: Claudex
Author: l3tchupkt

Issues

Claudex is an open-source AI coding agent with multi-provider support (OpenAI, NIM, Ollama, and more), featuring a smart routing system, Telegram integration, and full Claude Code tooling — all powered by any model you choose.

101stars

8forks

TypeScript

Installation

# Add to your Claude Code skills
git clone https://github.com/l3tchupkt/Claudex

Getting Started

Guides for using ai agents skills like Claudex.

Caveman: Cut Claude Token Use by 65%
How agent-side prompt compression works, when to use it, and when not to.
What is an AI Skills Marketplace?
Definitions, how marketplaces work, and how to choose between them in 2026.
Getting Started with AI Skills

Security ReportIssues

Last scanned: 10/17/2026

{
  "issues": [
    {
      "type": "npm-audit",
      "message": "@anthropic-ai/sdk: Claude SDK for TypeScript has Insecure Default File Permissions in Local Filesystem Memory Tool",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@mendable/firecrawl-js: Vulnerability found",
      "severity": "high"
    },
    {
      "type": "npm-audit",
      "message": "@modelcontextprotocol/sdk: MCP TypeScript SDK: OAuth client could send credentials to an authorization server chosen by the MCP server",
      "severity": "high"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/core: OpenTelemetry Core: Unbounded memory allocation in W3C Baggage propagation",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/exporter-logs-otlp-http: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/exporter-trace-otlp-grpc: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/otlp-exporter-base: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/otlp-grpc-exporter-base: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/otlp-transformer: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/resources: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/sdk-logs: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/sdk-metrics: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/sdk-trace-base: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "@opentelemetry/sdk-trace-node: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "axios: Axios has a NO_PROXY Hostname Normalization Bypass that Leads to SSRF",
      "severity": "high"
    },
    {
      "type": "npm-audit",
      "message": "diff: jsdiff has a Denial of Service vulnerability in parsePatch and applyPatch",
      "severity": "low"
    },
    {
      "type": "npm-audit",
      "message": "fflate: fflate unzipSync can enter an infinite loop when parsing malformed ZIP64 archives",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "gaxios: Vulnerability found",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "lodash-es: lodash vulnerable to Code Injection via `_.template` imports key names",
      "severity": "high"
    },
    {
      "type": "npm-audit",
      "message": "shell-quote: shell-quote quote() does not escape newlines in object .op values",
      "severity": "critical"
    },
    {
      "type": "npm-audit",
      "message": "undici: undici vulnerable to TLS certificate validation bypass via dropped requestTls in SOCKS5 ProxyAgent",
      "severity": "high"
    },
    {
      "type": "npm-audit",
      "message": "uuid: uuid: Missing buffer bounds check in v3/v5/v6 when buf is provided",
      "severity": "medium"
    },
    {
      "type": "npm-audit",
      "message": "ws: ws: Uninitialized memory disclosure",
      "severity": "high"
    }
  ],
  "status": "FAILED",
  "scannedAt": "2026-10-17T08:19:46.454Z",
  "npmAuditRan": true,
  "pipAuditRan": true,
  "promptInjectionRan": true
}

Compare with similar skills

See how Claudex compares with popular alternatives.

Claudex vs rea

See comparison

Claudex vs ecc

See comparison

Claudex vs hermes agent

See comparison

README.md

Frequently Asked Questions

What is Claudex?

Claudex is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by l3tchupkt. Claudex is an open-source AI coding agent with multi-provider support (OpenAI, NIM, Ollama, and more), featuring a smart routing system, Telegram integration, and full Claude Code tooling — all powered by any model you choose. It has 101 GitHub stars.

Is Claudex safe to use?

Claudex failed SkillsLLM's automated security scan, which flagged one or more high-severity issues. Review the Security Report section carefully before using it.

How do I install Claudex?

Clone the repository with "git clone https://github.com/l3tchupkt/Claudex" and add it to your Claude Code skills directory (see the Installation section above).

What programming language is Claudex written in?

Claudex is primarily written in TypeScript. It is open-source under l3tchupkt on GitHub, so you can review or fork the full source.

Are there alternatives to Claudex?

Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh Claudex against similar tools.

AI Courses · Agentic AI, LLM Engineering & MCP

Beginner to professional · First lesson free · Lifetime access

20% off

Comments (0)

to leave a comment.

No comments yet. Be the first to share your thoughts!

Related Skills

rea

by morluto

Reverse engineer anything with agents, from app behavior down to native binaries.

295,349

Binance-Claw Jixu

Claudex

Claudex is a fork of the Claude Code source that adds a full OpenAI-compatible provider shim, NVIDIA AI (NIM) support, a smart multi-provider router, a Telegram gateway, and local inference via Ollama and Atomic Chat. Every Claude Code tool works — bash, file ops, grep, glob, agents, MCP, tasks — powered by whatever model you choose.

Author: Lakshmikanthan K — github.com/l3tchupkt

Install

npm install -g @letchu_pkt/claudex

claudex

On first run, if no provider is configured, Claudex will prompt you to set one up via /provider. No Anthropic account required.

Fastest Start

Pick a provider, set three env vars, run.

NVIDIA AI — free key, best models

export CLAUDE_CODE_USE_NVIDIA=1
export NVIDIA_API_KEY=nvapi-your-key
export NVIDIA_MODEL=moonshotai/kimi-k2-instruct
claudex

Free key at build.nvidia.com.

OpenAI

export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_API_KEY=sk-your-key
export OPENAI_MODEL=gpt-4o
claudex

Google Gemini — free key

export CLAUDE_CODE_USE_GEMINI=1
export GEMINI_API_KEY=your-key
export GEMINI_MODEL=gemini-2.0-flash
claudex

Free key at aistudio.google.com/apikey.

Ollama — fully local, no key

ollama pull llama3.1:8b
export CLAUDE_CODE_USE_OPENAI=1
export OPENAI_BASE_URL=http://localhost:11434/v1
export OPENAI_MODEL=llama3.1:8b
claudex

Guides

Audience	Guide
New to terminals	Non-Technical Setup
Windows	Windows Quick Start
macOS / Linux	macOS / Linux Quick Start
Android (Termux)	Android Install
Source builds, profiles, diagnostics	Advanced Setup
Daily Ollama workflow	Playbook
Telegram bot	Telegram Gateway

Supported Providers

Provider	Env flag	Key var
NVIDIA AI (NIM)	`CLAUDE_CODE_USE_NVIDIA=1`	`NVIDIA_API_KEY`
OpenAI / any OpenAI-compatible	`CLAUDE_CODE_USE_OPENAI=1`	`OPENAI_API_KEY`
Google Gemini	`CLAUDE_CODE_USE_GEMINI=1`	`GEMINI_API_KEY`
GitHub Models	`CLAUDE_CODE_USE_GITHUB=1`	`GITHUB_TOKEN`
Amazon Bedrock	`CLAUDE_CODE_USE_BEDROCK=1`	AWS credentials
Google Vertex AI	`CLAUDE_CODE_USE_VERTEX=1`	GCP credentials
Microsoft Foundry	`CLAUDE_CODE_USE_FOUNDRY=1`	`ANTHROPIC_FOUNDRY_API_KEY`
Ollama (local)	`CLAUDE_CODE_USE_OPENAI=1` + localhost URL	none
Atomic Chat (Apple Silicon)	`CLAUDE_CODE_USE_OPENAI=1` + 127.0.0.1:1337	none
Anthropic (default)	none	`ANTHROPIC_API_KEY`

Any OpenAI-compatible endpoint works: DeepSeek, Groq, Mistral, Together AI, OpenRouter, LM Studio, Azure OpenAI, and more.

NVIDIA AI Models

Model	Best for
`moonshotai/kimi-k2-instruct`	Reasoning, coding (default)
`nvidia/llama-3.1-nemotron-ultra-253b-v1`	Flagship quality
`meta/llama-3.3-70b-instruct`	Balanced speed/quality
`meta/llama-3.1-8b-instruct`	Fast, lightweight
`deepseek-ai/deepseek-r1`	Deep reasoning
`qwen/qwen3-235b-a22b`	Large MoE
`mistralai/mistral-large-2-instruct`	Instruction following

Startup Themes

Set CLAUDEX_THEME to change the banner color scheme:

Theme	Colors
`sunset`	warm orange → rust (default)
`ocean`	deep teal → electric cyan
`aurora`	green → violet
`neon`	hot pink → electric blue
`mono`	white → grey

export CLAUDEX_THEME=ocean
claudex

Profile Launcher

Save a provider profile once, launch with one command:

# save a profile (also works via /provider inside the CLI)
bun run profile:init -- --provider nvidia --api-key nvapi-...
bun run profile:init -- --provider openai --api-key sk-...
bun run profile:init -- --provider ollama --model llama3.1:8b

# launch from saved profile
bun run dev:profile

# provider-specific launchers
bun run dev:nvidia
bun run dev:openai
bun run dev:ollama
bun run dev:gemini
bun run dev:codex
bun run dev:atomic-chat

Smart Router

Benchmarks all configured providers on startup and routes each request to the fastest, cheapest, healthiest option:

export ROUTER_MODE=smart
export ROUTER_STRATEGY=balanced   # latency | cost | balanced
claudex

Telegram Gateway

Use Claudex through a Telegram bot. Each user gets an isolated session.

# configure (one time)
claudex telegram setup --token 123456:ABC --provider nvidia

# allow yourself
claudex telegram permit 987654321

# start the gateway
claudex telegram start

Or manage from inside the CLI:

/telegram setup --token 123456:ABC --provider nvidia
/telegram permit 987654321
/telegram status

Full guide: telegram-gateway/README.md

What Works

All tools: Bash, FileRead, FileWrite, FileEdit, Glob, Grep, WebFetch, WebSearch, Agent, MCP, LSP, NotebookEdit, Tasks
Real-time token streaming
Multi-step tool chains
Base64 and URL image inputs (vision models)
Slash commands: /commit, /review, /compact, /diff, /doctor, /provider, /telegram, etc.
Sub-agents via AgentTool
Persistent memory

What's Different from Upstream

No Anthropic extended thinking (OpenAI models use different reasoning)
No prompt caching (Anthropic-specific)
No Anthropic beta headers
Token output defaults to 32K — models that cap lower are handled gracefully
First-run skips Anthropic login if a provider profile is saved

Web Search

WebSearch is disabled by default for non-Anthropic providers. Set a Firecrawl key to enable it:

export FIRECRAWL_API_KEY=your-key

Free tier includes 500 credits. With this set, WebSearch works for all providers and WebFetch handles JS-rendered pages.

How the Shim Works

Claude Code Tool System
        │
        ▼
  Anthropic SDK interface (duck-typed)
        │
        ▼
  openaiShim.ts  ◄── translates formats
        │
        ▼
  OpenAI Chat Completions API
        │
        ▼
  Any compatible model

Model Quality Reference

Model	Tool Calling	Code	Speed
GPT-4o	Excellent	Excellent	Fast
Kimi K2 (NVIDIA)	Excellent	Excellent	Fast
DeepSeek-V3	Great	Great	Fast
Gemini 2.0 Flash	Great	Good	Very Fast
Llama 3.3 70B	Good	Good	Medium
Mistral Large	Good	Good	Fast
GPT-4o-mini	Good	Good	Very Fast
Qwen 2.5 72B	Good	Good	Medium
Models < 7B	Limited	Limited	Very Fast

VS Code Extension

Install the bundled extension from vscode-extension/openclaude-vscode for one-click terminal launch and the Claudex Terminal Black theme.

Origin

Fork of the Claude Code source snapshot that became publicly accessible via an npm source map exposure on March 31, 2026. The original source is the property of Anthropic. This project is not affiliated with or endorsed by Anthropic.

License

Educational and research use. Original source subject to Anthropic's terms. Shim additions and new provider integrations are public domain.