assayer 1.0.0

Run a prompt across multiple LLMs and compare outputs side by side in the terminal.

Assayer

Send a prompt to multiple language models in parallel and compare their outputs in the terminal. Useful for evaluating which model handles a given task better, measuring semantic similarity between responses, or running an LLM-as-judge evaluation - without leaving the shell.

Installation

pip install assayer

Similarity scoring requires the optional score extra:

pip install "assayer[score]"

Python 3.11 or newer is required.

Supported Providers

• OpenAI: All GPT models.
• Anthropic: Claude models (Opus 4.7, Sonnet 4.6, Haiku 4.5, and earlier).
• Google Gemini: Gemini 2.x and 3.x models.
• Ollama: Local models running on your machine.

Configuration

Assayer looks for API keys in environment variables or a configuration file at ~/.assayer/config.json.

Environment Variables

export OPENAI_API_KEY="your-key"
export ANTHROPIC_API_KEY="your-key"
export GEMINI_API_KEY="your-key"

Configuration File

{
  "OPENAI_API_KEY": "sk-...",
  "ANTHROPIC_API_KEY": "sk-ant-...",
  "GEMINI_API_KEY": "..."
}

Use assayer models check to verify your configuration.

Quickstart

assayer run "Explain recursion in one sentence." --models gpt-4o,claude-haiku-4-5-20251001

Commands

run

assayer run "prompt" --models gpt-4o,claude-sonnet-4-5
assayer run --prompt-file prompt.txt --models gpt-4o,ollama/llama3.2
assayer run "prompt" --models gpt-4o,claude-sonnet-4-5 --score
assayer run "prompt" --models gpt-4o,claude-sonnet-4-5 --judge gpt-4o --judge-criteria "clarity,brevity"
assayer run "prompt" --models gpt-4o,claude-sonnet-4-5 --output results.json
assayer run "prompt" --models gpt-4o,claude-sonnet-4-5 --output results.csv
assayer run "prompt with {var}" --models gpt-4o --var key=value

Flag	Description
--models	Comma-separated model identifiers (required)
--prompt-file	Path to a .txt file instead of an inline prompt
--var	KEY=VALUE template variable, repeatable
--system	System prompt applied to all models
--temperature	Sampling temperature
--max-tokens	Maximum output tokens
--score	Show pairwise similarity matrix
--judge	Model to use as judge
--judge-criteria	Comma-separated criteria for the judge
--output	Save results to .json or .csv
--timeout	Per-model timeout in seconds (default: 30)

models

assayer models list               # list all supported model identifiers
assayer models check              # check which API keys are configured
assayer models check ollama       # check if Ollama is running and list local models

config

assayer config set OPENAI_API_KEY sk-...
assayer config show

Keys are saved to ~/.assayer/config.json. Environment variables take precedence.

Providers

OpenAI

export OPENAI_API_KEY=sk-...

Supported models: gpt-5.5, gpt-5.5-pro, gpt-5.4, gpt-5.4-pro, gpt-5.4-mini, gpt-5.4-nano, gpt-5.2, gpt-5, gpt-5-mini, gpt-5-nano, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-4o, gpt-4o-mini, o3, o3-mini, o4-mini

Anthropic

export ANTHROPIC_API_KEY=sk-ant-...

Supported models: claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5-20251001, claude-opus-4-6, claude-sonnet-4-5, claude-opus-4-5

Google Gemini

export GEMINI_API_KEY=...

Supported models: gemini-3.1-pro-preview, gemini-3.1-flash-lite, gemini-3-flash-preview, gemini-2.5-pro, gemini-2.5-flash, gemini-2.5-flash-lite, gemini-2.0-flash, gemini-2.0-flash-lite

Ollama (local)

No API key needed. Start Ollama and use the ollama/ prefix:

ollama serve
assayer run "prompt" --models ollama/llama4-scout,ollama/llama3.2,ollama/qwen3

Scoring

--score embeds all outputs using all-MiniLM-L6-v2 (runs locally, no API call) and displays a pairwise cosine similarity matrix. Values range from 0 (unrelated) to 1 (identical meaning).

LLM-as-judge

--judge <model> sends all outputs to the specified model and asks it to pick a winner. Use --judge-criteria to focus the evaluation:

assayer run "Write a sorting algorithm." \
  --models gpt-4o,claude-sonnet-4-5 \
  --judge gpt-4o \
  --judge-criteria "correctness,readability"

If the judge call fails, a warning is printed to stderr and the run continues normally.

Export

--output results.json saves full results as JSON. --output results.csv saves as CSV. The file format is determined by the extension.

Contributing

Contributions are welcome. See CONTRIBUTING.md for setup instructions, code style, and the PR process.

License

MIT - see LICENSE for details.