pitchai-gpt (lib_agents)

Reusable GPT helpers and the Fast Streaming Agent runtime that power AFASAsk deployments. The package bundles two major pieces of functionality:

• lib_agents.gpt / lib_agents.gpt_root: async GPT helpers with smart prompt loading, Gemini/OpenRouter/Groq/Azure routing, and streaming support.
• lib_agents.fast_streaming_agent_root: the Fast Streaming Agent runtime for dataframe-aware copilots, including dataframe loading, semantic-layer hooks, and templated UI helpers.

Both modules are designed to be shared between repositories (e.g. AFASAsk and AI Price Crawler) so they can evolve in one place and be installed through PyPI. The distribution on PyPI is called pitchai-gpt, but it exposes the lib_agents Python package to avoid breaking existing imports.

Installation

uv pip install pitchai-gpt
# or
pip install pitchai-gpt

Environment variables

The GPT helpers expect API keys/endpoints to be provided through environment variables (or by passing api_key= directly when instantiating GPT). The most commonly used variables are:

The streaming agent runtime reads configuration from YAML/JSON prompt files, optionally loads .env files via python-dotenv, and integrates with Pandas/Polars. Make sure your application provides any referenced config modules (e.g. configs.server_config, libs.lib_utils).

Usage

from lib_agents.gpt.gpt import GPT

gpt = GPT(
    sys_msg="You are a concise assistant.",
    user_msg="Say READY",
    model="gemini-2.0-flash",
)
response, *_ = asyncio.run(gpt.run())
print(response)

For streaming agents see lib_agents.fast_streaming_agent_root.main.FastStreamingAgent and the demos under apps/web_app.

Development

This project uses Hatch via uv for building and publishing. See docs/uv_pypi_publish.md in the monorepo for a step-by-step release checklist.