text-gen

text-gen is a Bun + TypeScript CLI for provider-agnostic text generation with:

• OpenAI (text-gen openai)
• Gemini (text-gen gemini)
• Claude (text-gen claude)

It supports direct/file/stdin prompts, template interpolation, JSON output, and writing generated text to disk.

Install and Setup (Bun)

Requirements

• Bun >=1.1.0

Install dependencies

bun install

Run in development

bun run dev -- openai --model gpt-4o-mini --user "Say hello"

Build

bun run build

API Keys

Set provider keys in your shell environment:

export OPENAI_API_KEY="your_openai_key"
export GEMINI_API_KEY="your_gemini_key"
# or use GOOGLE_API_KEY for Gemini
export GOOGLE_API_KEY="your_google_key"
export ANTHROPIC_API_KEY="your_anthropic_key"

Provider key resolution:

• openai: OPENAI_API_KEY
• gemini: GEMINI_API_KEY or GOOGLE_API_KEY
• claude: ANTHROPIC_API_KEY

Command Reference

Global

text-gen --help
text-gen -h
text-gen --version
text-gen -v

Providers

text-gen openai [options]
text-gen gemini [options]
text-gen claude [options]

Provider help:

text-gen openai --help
text-gen gemini --help
text-gen claude --help

Full Flag Reference

All provider commands support:

• --model <id>: optional model identifier (defaults shown below)
• --temperature <number>: sampling temperature (validated to 0..2)
• --max-tokens <integer>: max output tokens (integer, validated >0)
• --top-p <number>: nucleus sampling (validated to 0..1)
• --system <text>: inline system prompt
• --user <text>: inline user prompt
• --system-file <path>: read system prompt from file
• --user-file <path>: read user prompt from file
• --var <key=value>: template variable, repeatable
• --json: print structured JSON payload
• --output <path>: write generated text to a file
• -h, --help: command help

Rules:

• If omitted, --model defaults to:
  • openai: gpt-5.2-chat-latest
  • gemini: gemini-pro-latest
  • claude: claude-sonnet-4-20250514
• --var can be repeated; most other flags can only appear once.
• You cannot combine --system with --system-file.
• You cannot combine --user with --user-file.

Prompt Input Modes

User prompt resolution order:

1. --user <text>
2. --user-file <path>
3. stdin fallback (only if no --user or --user-file is provided)

System prompt can come from:

• --system <text>, or
• --system-file <path>

Direct text

text-gen openai --model gpt-4o-mini --user "Summarize this release note."

File input

text-gen gemini --model gemini-2.5-flash --system-file prompts/system.txt --user-file prompts/user.txt

stdin fallback

echo "Write a 1-line tagline for a coffee shop." | text-gen claude --model claude-3-7-sonnet

Template Variables

Use {{key}} placeholders in prompt text and pass values via repeatable --var key=value.

Example template:

Write a {{tone}} summary about {{topic}}.

CLI usage:

text-gen openai \
  --model gpt-4o-mini \
  --user "Write a {{tone}} summary about {{topic}}." \
  --var tone=concise \
  --var topic=observability

File + variables:

text-gen gemini \
  --model gemini-2.5-flash \
  --user-file prompts/template.txt \
  --var product="text-gen" \
  --var audience=developers

Output Modes

Default (plain text)

Without --json, stdout is only the generated text.

JSON mode (--json)

With --json, stdout is:

{
  "success": true,
  "provider": "openai",
  "model": "gpt-4o-mini",
  "text": "Hello world",
  "usage": {
    "inputTokens": 1,
    "outputTokens": 2,
    "totalTokens": 3
  },
  "finish_reason": "stop"
}

Notes:

• usage and finish_reason can be null depending on provider response.
• raw provider output is not printed by the CLI payload.

--output <path> behavior

• Always writes generated text to the target file.
• Still prints to stdout (plain text without --json, JSON payload with --json).

Examples

OpenAI

text-gen openai \
  --model gpt-4o-mini \
  --system "You are concise." \
  --user "Explain retries in distributed systems in 3 bullets." \
  --temperature 0.2

Gemini

text-gen gemini \
  --model gemini-2.5-flash \
  --user "Generate 5 release-note headlines for a CLI tool." \
  --top-p 0.9 \
  --max-tokens 120

Claude

text-gen claude \
  --model claude-3-7-sonnet \
  --user "Rewrite this paragraph for a technical audience." \
  --json

stdin + output file

cat prompts/user.txt | text-gen openai --model gpt-4o-mini --output generated.txt

file prompt + JSON

text-gen claude --model claude-3-7-sonnet --user-file prompts/user.txt --json

Troubleshooting

Missing API keys

Examples:

• OpenAI: Missing API key for OpenAI. Fix: set OPENAI_API_KEY...
• Gemini: Missing API key for Gemini. Fix: set GEMINI_API_KEY or GOOGLE_API_KEY...
• Claude: Missing API key for Claude. Fix: set ANTHROPIC_API_KEY...

Fix by exporting the required key(s) before running.

Conflicting prompt arguments

If both inline and file are passed for the same field:

• --user + --user-file
• --system + --system-file

CLI fails with a conflict error. Use one source per field.

Unresolved template variables

If placeholders remain after interpolation, CLI fails, for example:

Unresolved template variables in user prompt: {{topic}}. Provide values via --var key=value.

Add all required --var key=value entries.

Flag parsing/validation failures

Examples:

• Empty model value still fails, e.g. --model without a value.
• Bad number: Invalid --temperature value "nan". Expected a number.
• Bad integer: Invalid --max-tokens value "12.5". Expected an integer.
• Range checks: temperature must be 0..2, topP must be 0..1, and maxTokens must be >0.

Missing user prompt

If no --user, no --user-file, and no stdin input:

Missing user prompt. Provide --user, --user-file, or pipe content via stdin...

Development, Scripts, and Tests

• bun run dev: run CLI entrypoint in development
• bun run build: build distributable files to dist/
• bun run typecheck: TypeScript type checks (tsc --noEmit)
• bun run test: run test suite (bun test)
• bun run check: run typecheck then test

Local verification:

bun run typecheck
bun test