AIP

AIP is a unified command-line playground for model chat, comparisons, agents, experiments, prompt templates, vector search, RAG, and optional web interfaces.

Install

Python 3.11 or newer is required.

uv tool install .
# or, for development
uv sync --extra dev

The installed command and import package are aip. The PyPI distribution is ai-playground, preserving the existing project's upgrade path:

pip install --upgrade ai-playground

Quickstart

Ollama is the default provider:

aip model list
aip chat "Explain retrieval-augmented generation in two sentences"
aip model chat llama3:8b "Write a Python hello-world program"
aip compare "Explain RAG" --models ollama:llama3,litellm:gpt-4o-mini
aip shell

To use OpenAI, store the API key and select defaults:

aip config set openai.api_key sk-...
aip config set models.default_provider openai
aip config set models.default_model gpt-4o-mini
aip chat "Hello"

Configuration is saved in ~/.aip/config.yaml. Environment variables override saved values using nested names such as AIP_MODELS__DEFAULT_MODEL and AIP_OPENAI__API_KEY. API keys are redacted by aip config show.

Azure OpenAI is also supported with the provider name azure:

aip config set azure_openai.endpoint https://example.openai.azure.com
aip config set azure_openai.api_key your-key
aip model chat deployment-name "Hello" --provider azure

Diagnostics and development

aip version
aip info
aip doctor
uv run pytest
uv run ruff check .
uv run mypy
uv build

aip doctor returns a failing status when the selected default provider is not available, making it useful in setup scripts and CI.

Research workflows

aip experiment start "model comparison"
aip chat "Explain attention"
aip experiment end
aip templates use summarize --file report.txt
aip usage show --by-model

For retrieval-augmented generation, install a vector-store extra and ingest documents:

pip install 'ai-playground[vectordb-chroma]'
aip vectordb create chroma docs
aip vectordb add docs --file README.md
aip vectordb rag docs "How do I install AIP?"

Plugins and optional UIs

AIP supports built-in, installed entry-point, and local-file plugins. Browse and run them without coupling the core CLI to optional frameworks:

aip plugin list
aip marketplace search ui
aip plugin info streamlit
uv tool install 'ai-playground[streamlit]'
aip plugin run streamlit serve

See the plugin API guide and the hello-aip reference plugin.

Notebooks

AIP ships five Marimo notebooks that run with the same packages as the CLI. Install the extra, materialize them locally, then edit or run:

uv tool install 'ai-playground[marimo]'
aip notebooks init
aip notebooks edit hello
aip notebooks run chat

hello is a minimal sanity check, chat is a chat playground over any aip provider, compare lays model responses out side by side, usage renders read-only usage and experiment dashboards, and rag is a retrieval playground over a vector database. You can also serve the gallery with aip serve marimo. See docs/notebooks.md.

The complete guides and cookbook live in docs. See CONTRIBUTING.md for the development quality gate.