quickstart

quickstart

Single-agent ReAct demo using an Anthropic-compatible endpoint (MiniMax by default).

中文文档

Quick start (dev environment)

# 1. Install dependencies
uv sync

# 2. Configure credentials
cp examples/quickstart/.env.example examples/quickstart/.env
# Edit .env — fill in LLM_API_KEY, LLM_API_BASE, LLM_MODEL

# 3. Run via the built-in CLI
openagents run examples/quickstart/agent.json --input "hello"

Other output formats:

# JSONL event stream (for pipelines)
openagents run examples/quickstart/agent.json --input "hello" --format events

# Full RunResult as JSON
openagents run examples/quickstart/agent.json --input "hello" --format json --no-stream

# Interactive multi-turn chat
openagents chat examples/quickstart/agent.json

Legacy script

run_demo.py remains for historical reference; it's a one-shot wrapper that loads .env, constructs a Runtime, and issues two hand-coded runs. New code should prefer openagents run / openagents chat.

Testing

The quickstart agent is exercised by the CLI smoke tests and the openagents run integration tests, all of which use the mock provider and require no real API key:

uv run pytest -q tests/integration/test_cli_smoke.py
uv run pytest -q tests/unit/cli/commands/test_run.py

Environment variables

Name	Required	Notes
LLM_API_KEY	yes	OpenAI-compatible key.
LLM_API_BASE	yes	Base URL of the provider (e.g. https://api.minimax.chat/anthropic).
LLM_MODEL	yes	Model name (e.g. abab6.5-chat).