Blazing Agents Python SDK

Build production agents in Python with a typed client for the Blazing Agents API.

Documentation · PyPI

The official Python SDK provides synchronous and asynchronous clients for the Blazing Agents /v1 API. It supports CPython 3.11 and newer.

Features

• Typed Pydantic request and response models.
• Matching synchronous and asynchronous APIs.
• Agent, Workspace, Skill, Provider, Prompt, Memory, Session, Artifact, Task, usage, and Tenant management.
• Chat, text, and structured-object generation streams.
• Lazy pagination and binary transfers.
• Request correlation with configurable timeouts and observability.

Installation

pip install blazing-agents

Quick start

Create a Tenant API key in the Blazing Agents dashboard, then pass it to the client or set BLAZING_AGENTS_API_KEY.

from blazing_agents import BlazingAgents


with BlazingAgents(api_key="ba_...") as client:
    result = client.completion(
        agent_id="ag_...",
        prompt="Write a friendly welcome message.",
    )
    print(str(result))

Use AsyncBlazingAgents for asynchronous applications; it exposes the same resources and generation methods.

Documentation

Read the Python SDK documentation for authentication, resource guides, generation and streaming, error handling, and the complete API reference.

Thinking levels

Configure thinking_level on Agent create or update. Omit it on update to preserve the current selection; pass None for Provider default. Explicit levels are strings, including custom values for Models with unknown capabilities. Agent and Agent Version responses expose thinking_level, and restoring a Version restores its level too. The async client provides the same methods.

capabilities = client.providers.get_thinking_levels(provider_id, model="gpt-5")
# capabilities.known distinguishes unknown metadata from known choices.
agent = client.agents.create(
    name="Reasoner",
    provider_id=provider_id,
    model="openai/gpt-5",
    thinking_level="high",
)
client.agents.update(agent.id, thinking_level=None)

Development

uv sync --locked
uv run ruff check .
uv run ruff format --check .
uv run python scripts/run_typechecks.py
uv run python scripts/run_tests.py

License

MIT

Interactive resend

Successful interactive exchanges are saved together. Failed or canceled execution leaves saved history unchanged, including the previous answer during regeneration; executed usage and Tool effects remain. Retain submitted text/images until success and resend edited or unchanged input through ordinary chat with a fresh message ID. Stop requests cancellation; a lost response can hide a saved exchange. Reuse the returned Session ID and load history normally on return. No outcome polling or automatic generation retry is needed. See the chatbot guide and working examples.