chatgpt: Chat GPT console client in Golang

English | 中文

A Golang console client for ChatGPT (https://chat.openai.com) using GPT

Features

• Chat Completions API with streaming output and multi-turn conversation memory
• Selectable model (--model / OPENAI_MODEL) and custom base URL (--base-url / OPENAI_BASE_URL) for Azure OpenAI or other OpenAI-compatible services
• Non-interactive and piped input (-p)
• /save and /load conversations (JSON), /system, /reset
• Token usage and estimated cost, colored code blocks, Ctrl+C to interrupt, request timeout

Install

go get -u -x github.com/kkdai/chatgpt

Request for API from OpenAI

Request your OpenAPI key in https://platform.openai.com/api-keys

Usage

export API_KEY=YOUR_KEY   # or OPENAI_API_KEY
chatgpt

Options:

-m, --model string     model to use (default "gpt-4o", env OPENAI_MODEL)
-s, --system string    system prompt
-p, --prompt string    send one prompt and exit
    --base-url string  OpenAI-compatible API base URL (env OPENAI_BASE_URL)
    --timeout duration maximum time for one answer (default 5m0s, 0 for no limit)

For one-shot or piped use, pass a prompt with -p; piped text is appended to that prompt. Piped input without -p is sent as a single prompt:

cat file.txt | chatgpt -p "Summarize this"
echo "What is Go?" | chatgpt

The conversation keeps its history across interactive questions. In-session commands: /system <prompt>, /reset, /save <file>, /load <file>, /image <file>, /help, quit / exit. Conversation files use JSON; /save refuses to overwrite an existing file unless /save --force <file> is used. Press Ctrl+C while an answer is streaming to interrupt it.

When the API reports token usage, it is printed to stderr along with an estimated cost for supported GPT-4o and GPT-4.1 models. Estimates use published per-token rates and may not match provider-specific pricing. Code fences are colored when output is a terminal; redirected output remains plain text.

Snapshot

Contribute

Please open up an issue on GitHub before you put a lot efforts on pull request. The code submitting to PR must be filtered with gofmt

License

This package is licensed under MIT license. See LICENSE for details.

Reasoning, vision and structured output

• --reasoning-effort low|medium|high (env OPENAI_REASONING_EFFORT) for o-series models.
• /image <file> attaches a local image to the next question (vision-capable models).
• --json-schema '<schema>' or --json-schema @schema.json forces output to match a JSON schema.
• /draw <prompt> generates an image (model from OPENAI_IMAGE_MODEL, default gpt-image-1) and saves it as draw-<time>.png in the current directory.