ChogGPT

README

ChogGPT 🤖

ChogGPT is a cross-platform Command-Line Interface (CLI) tool that lets you talk to AI models — chat, generate code, get shell commands, search the web, generate images — all from your terminal.

This repository is a fork of aandrew-me/tgpt, rebranded and maintained as ChogGPT. All credit for the original design and implementation goes to the upstream project and its contributors — see Credits below.

---

Table of contents

• Features
• Installation
• Quick start
• Supported AI providers
• Image generation providers
• Tool calling & MCP
• Configuration
• Proxy support
• Building from source
• Updating
• Uninstalling
• Credits & License

---

Features

• 🧠 Chat with 20+ AI providers — many of them completely free, no API key required
• 💬 Interactive modes: normal, multi-line, shell, and web-search-driven
• 💻 Generate and (optionally) execute shell commands directly (-s)
• 🧑‍💻 Generate code (-c)
• 🖼️ Generate images from text (--img), via several free/paid providers
• 🔎 Web search built in, with exa or google as the search backend (-f)
• 🛠️ Built-in tool calling (file read/write, grep, glob, shell exec, web fetch, web search) and full MCP (Model Context Protocol) server support
• 🔁 Automatic provider fallback/rotation if one provider fails (--rotate)
• 🌐 HTTP/SOCKS5 proxy support
• 📦 Single static binary, no runtime dependencies — works on Linux, macOS, Windows, FreeBSD, NetBSD, OpenBSD, and multiple architectures (amd64, arm64, arm, 386, including softfloat variants)

---

Installation ⏬

Build from source (recommended for this fork)

Since this is a personal fork, the package-manager listings that exist for upstream tgpt (Homebrew, Pacman, Chocolatey, FreeBSD ports) are not published for ChogGPT. The two reliable ways to get ChogGPT are:

1. go install (requires Go ≥ 1.25):

go install github.com/0xCaffeeNads/ChogGPT/v2@latest

Make sure your Go bin directory ($(go env GOPATH)/bin) is on your PATH.

2. Clone and build manually:

git clone https://github.com/0xCaffeeNads/ChogGPT.git
cd ChogGPT
CGO_ENABLED=0 go build -trimpath -ldflags="-s -w" -o choggpt
sudo mv choggpt /usr/local/bin/   # optional, for global access

⚠️ If you're building cross-platform (e.g. from Termux/Android for a Linux/Windows/macOS target), always set GOOS/GOARCH explicitly, for example:

CGO_ENABLED=0 GOOS=linux GOARCH=amd64 go build -trimpath -ldflags="-s -w" -o choggpt-linux-amd64

Omitting GOOS lets Go default to the host platform, which on Android/Termux can pull in mobile-only code paths that fail to compile.

Prebuilt binaries (GitHub Releases)

This repo's .github/workflows/main.yml CI workflow builds binaries for every supported platform and architecture (Linux, Windows, macOS, FreeBSD, NetBSD, OpenBSD × amd64/arm64/arm/386, plus softfloat variants) and attaches them as a draft GitHub Release whenever it runs.

• It runs automatically on pushes to the release branch.
• It can also be triggered manually from Actions → Build Go Project and Create Release → Run workflow, where you can pick a specific os and arch to build just that one binary instead of the full matrix (or leave both as all).

Once a release is published, download the binary matching your OS/architecture from the Releases page, rename it to choggpt (optional), make it executable, and place it somewhere on your PATH:

chmod +x choggpt-linux-amd64
sudo mv choggpt-linux-amd64 /usr/local/bin/choggpt

Windows 🪟

irm https://raw.githubusercontent.com/0xCaffeeNads/ChogGPT/main/install-win.ps1 | iex

Uninstall with:

iex "& { $(irm https://raw.githubusercontent.com/0xCaffeeNads/ChogGPT/main/install-win.ps1) } -Uninstall"

---

Quick start

# Ask a question (uses the default provider, opencode — free, no key needed)
choggpt "What is internet?"

# Multi-line interactive mode
choggpt -m

# Generate & confirm-run a shell command
choggpt -s "How do I find large files in this directory?"

# Generate code
choggpt -c "Write a function in Go that reverses a string"

# Use a specific provider
choggpt --provider groq "What is 1+1"

# Generate an image
choggpt --img "a cat astronaut, digital art"
choggpt --img --out ~/cat.jpg --width 512 --height 512 "a cat astronaut"

# Web search
choggpt -f "latest news about AI"

# Pipe input in
cat main.go | choggpt "Explain this code"

Run choggpt -h for the full flag reference, or see md/usage.md.

---

Supported AI providers 🤖

The default provider is opencode (free, no API key needed). Switch providers with --provider <name> or the AI_PROVIDER environment variable.

Provider	Cost	Key environment variables	Notes
opencode (default)	Free	OPENCODE_API_KEY, OPENCODE_MODEL, OPENCODE_URL	Uses opencode.ai/zen API, deepseek-v4-flash-free by default
anyapi	Free (100k tokens/day)	ANYAPI_API_KEY, ANYAPI_MODEL	Multi-model gateway (OpenAI, Google, DeepSeek, and more); also supports image generation
pollinations	Free	POLLINATIONS_API_KEY, POLLINATIONS_MODEL	Many free text models
powerbrain	Free	—	Uses gpt-5 by default, no key required
isou	Free	—	DeepSeek-chat backed by SearXNG web search
koboldai	Free	—	koboldcpp / HF Space "Tiefighter-13B"
fx	Free	—	fx.sh gateway, default zai/glm-5.2
aihorde	Free (community-run)	AIHORDE_API_KEY, AIHORDE_MODEL	Volunteer-powered compute; also does image generation
groq	Free API key	GROQ_API_KEY, GROQ_MODEL	Fast inference, many models
gemini	Free API key	GEMINI_API_KEY, GEMINI_MODEL	Default gemini-2.0-flash
openrouter	Free tier	OPENROUTER_API_KEY, OPENROUTER_MODEL, OPENROUTER_URL	Default openrouter/free
ollama	Free (local)	—	Runs local models via a local Ollama install
ollamacloud	Requires API key	OLLAMA_API_KEY, OLLAMA_MODEL	Default gpt-oss:120b
deepseekweb	Free (browser token)	DEEPSEEK_WEB_TOKEN / --key	Uses your chat.deepseek.com session; needs node/bun/deno in PATH to solve PoW
openai	Paid API key	OPENAI_API_KEY, CEREBRAS_API_KEY, OPENAI_MODEL	Any OpenAI-compatible model; custom URL via --url
deepseek	Paid API key	DEEPSEEK_API_KEY, DEEPSEEK_MODEL	Default DeepSeek-V4-Flash
atlascloud	Paid API key	ATLASCLOUD_API_KEY, ATLASCLOUD_MODEL, ATLASCLOUD_URL	OpenAI-compatible, default qwen/qwen3.8-max
minimax	Paid API key	MINIMAX_API_KEY, MINIMAX_MODEL	Default MiniMax-M2.7
litellm	Depends (self-hosted)	LITELLM_API_KEY, LITELLM_MODEL, LITELLM_URL	Gateway to 100+ providers, requires --model
omniroute	Self-hosted router	OMNIROUTE_API_KEY, OMNIROUTE_MODEL, OMNIROUTE_URL	Default http://localhost:20128/v1, model auto

Full details for every provider: md/providers.md

Image generation providers 🖼️

Provider	Cost	Notes
pollinations	Free	Models: flux, turbo
magicstudio	Free	Very fast
aihorde	Free (community-run)	Volunteer-powered
anyapi	Requires API key	Many models, incl. google/gemini-2.5-flash-image

Set the image provider with IMG_PROVIDER (separate from the chat AI_PROVIDER).

---

Tool calling & MCP

ChogGPT can let the model call tools while answering (-t / --tools), and can connect to MCP (Model Context Protocol) servers to add more tools at runtime. Not every provider supports tool calling — unsupported ones simply ignore the tool definitions.

Built-in tools:

web_search_exa, web_search_firecrawl, read_directory, read_file, execute_command, web_fetch, write_file, edit_file, grep, glob

choggpt -t "What files are in the current directory?"
choggpt -t web_search_exa,read_file "Search and read the specified file"
choggpt --mcp "Use MCP tools from auto-detected mcp_config.json"
choggpt --mcp-server "npx -y @modelcontextprotocol/server-filesystem /path/to/dir" "List the files"
choggpt --mcp-add       # interactively configure a new MCP server
choggpt --mcp-remove    # interactively remove one

Example mcp_config.json:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/dir"],
      "env": ["LOG_LEVEL=info"]
    },
    "firecrawl": {
      "url": "https://mcp.firecrawl.dev/v2/mcp",
      "headers": { "Authorization": "Bearer YOUR_FIRECRAWL_API_KEY" }
    }
  }
}

---

Configuration

Every provider's API key/model/URL can be set via environment variables (see the provider table above), or persisted in a config file:

• ~/.config/choggpt/config.conf on Linux/macOS (%USERPROFILE%\.config\choggpt on Windows)
• ./config.conf in the current directory (takes priority over the home directory config)

Example:

AI_PROVIDER=groq
GROQ_API_KEY=gsk_xxxx
OPENCODE_MODEL=mimo-v2.5-free
AI_ROTATE_PROVIDERS=deepseek,groq,anyapi,opencode
HTTP_PROXY=http://localhost:8080

For web search (-f), the default backend is exa (works without a key, rate-limited; supports EXA_API_KEY). To use Google instead, set SEARCH_PROVIDER=google and see SEARCH_SETUP.md for CHOGGPT_GOOGLE_API_KEY / CHOGGPT_GOOGLE_SEARCH_ENGINE_ID setup.

---

Proxy support

1. Environment variable — http_proxy / HTTP_PROXY:

• HTTP: http://ip:port
• HTTP with auth: http://user:pass@ip:port
• SOCKS5: socks5://ip:port
• SOCKS5 with auth: socks5://user:pass@ip:port

2. Config file — put the proxy URL in ./proxy.txt or ~/.config/choggpt/proxy.txt:

http://127.0.0.1:8080

---

Building from source

git clone https://github.com/0xCaffeeNads/ChogGPT.git
cd ChogGPT
go build -trimpath -ldflags="-s -w" -o choggpt

Or use build.sh to cross-compile every supported platform/architecture locally, matching what CI produces.

CI configuration lives in .github/workflows/main.yml and supports manual, selective builds via workflow_dispatch inputs (os, arch) — see Prebuilt binaries above.

---

Updating ⬆️

If installed via the install script:

choggpt -u

(May require admin privileges depending on install location.)

If installed via go install, re-run:

go install github.com/0xCaffeeNads/ChogGPT/v2@latest

---

Uninstalling

sudo rm $(which choggpt)

The configuration directory is ~/.config/choggpt on Linux (~/Library/Application Support/choggpt on macOS) — remove it too if you want a clean uninstall.

---

Credits & License

ChogGPT is a fork of tgpt by aandrew-me and its contributors. All original design, architecture, and provider integrations are their work — this fork exists for rebranding/customization purposes.

Licensed under the GNU General Public License v3.0, same as upstream.