eager-tools-openai

OpenAI adapter for eager tool calling — wraps chat.completions.create(stream=True) and dispatches each tool the instant its block seals, overlapping tool execution with ongoing LLM generation.

For the mechanism and runtime contract, see the top-level METHOD.md.

Status

Alpha / scaffold. Public API shape is locked; implementation bodies arrive in Move 3 (CloudThinker port). See NEXT.md.

Install (once published)

pip install eager-tools-openai

Usage

from openai import AsyncOpenAI
from eager_tools_openai import OpenAIEagerStream

client = AsyncOpenAI()

source = await client.chat.completions.create(
    model="gpt-4.1",
    tools=[...],
    messages=[...],
    stream=True,
)
runner = OpenAIEagerStream(source, tools=my_tools)

async for event in runner.events():
    print(event.kind, event.tool_call)

async for call, result in runner.results():
    print(call.name, "→", result)

Supported SDK surfaces

• chat.completions.create(stream=True) with tools=[...] — v0.1 target.
• Responses API (responses.stream(...)) — deferred; contributions welcome.

Supported SDK versions

• openai>=1.50.

License

MIT. See top-level LICENSE.