bp-agent 0.5.0

Minimal task execution agent framework

bp-agent

For the developer's vision and thoughts behind this project, see HUMANS.md.

Minimal task execution agent framework with multi-provider LLM support.

Features

• Multi-provider: Gemini, Codex (OpenAI), Opus with automatic key rotation
• Streaming: Real-time token streaming for chat responses
• Tool system: Built-in tools (bash, read_file, write_file, list_dir) + custom tools
• Subagents: Spawn worker agents for parallel task execution
• Chat mode: Multi-turn conversation with tool support
• Task queue: Persistent JSON-based task scheduling

Install

pip install bp-agent

Quick Start

# Set API key
export GEMINI_API_KEY=your_key

# Interactive chat (streaming)
bp-chat

# Chat with a specific provider
bp-chat --provider codex
bp-chat --provider opus --model my-model

Providers

Provider	Env Vars	Models
Gemini (default)	GEMINI_API_KEY	gemini-3-flash-preview, gemini-3-pro-preview
Codex	CODEX_API_KEY or ~/.codex/auth.json	gpt-5.2-codex, gpt-5.1-codex-mini, ...
Opus	OPUS_API_KEY + OPUS_BASE_URL	configurable

Multiple keys supported via GEMINI_API_KEY_2, _3, etc. or comma-separated GEMINI_API_KEYS.

Python API

from bp_agent import Agent, AgentConfig

# Task execution
agent = Agent("my-agent")
result = agent.execute("List all Python files in src/")
print(result.output)

# Multi-turn chat
response = agent.chat("What files are in this directory?")
print(response)

# Streaming chat
for chunk in agent.chat_stream("Explain this codebase"):
    print(chunk, end="", flush=True)

# Custom tools
from bp_agent.tools import ToolSchema
agent.add_tool("greet", lambda name: f"Hello, {name}!", ToolSchema(
    name="greet",
    description="Greet someone",
    parameters={"type": "object", "properties": {"name": {"type": "string"}}, "required": ["name"]},
))

License

GPL-3.0