openapi-batch 0.1.0

Reliable, async batch execution of LLM requests across providers.

openapi-batch

A small Python library that makes batch LLM requests reliable and easy across providers.

It handles:

• async submission (non-blocking by default)
• retries and partial failures
• durable progress tracking (SQLite)
• provider differences (OpenAI, Gemini, others)
• callbacks when results are ready

You submit items → you get a job handle → results arrive later.

---

Install

pip install openapi-batch

Optional provider extras:

pip install openapi-batch[openai]
pip install openapi-batch[gemini]
pip install openapi-batch[openai,gemini]

---

Quick example

Async batch with callback (recommended)

from openapi.batch import BatchClient

def on_done(job):
    print("status:", job.status())
    print(job.results_dict())

client = BatchClient(
    provider="openai",
    api_key="...",
    default_model="gpt-4o-mini",
)

job = client.map(
    mode="native",   # native | emulated | auto
    items=[
        {"item_id": "a", "input": {"prompt": "Return OK"}},
        {"item_id": "b", "input": {"prompt": "Return YES"}},
    ],
    on_complete=on_done,
)

print("submitted:", job.job_id)

Submission returns immediately. Processing happens in the background.

---

Blocking usage (for scripts/tests)

job = client.map(items=items, async_submit=False)
job.wait()

results = job.results_dict()

---

Concepts (brief)

Job

A batch execution with a stable job_id. Stored durably in SQLite.

Item

One logical request. Identified by item_id. If you don’t provide one, a deterministic ID is generated.

Modes

• emulated – concurrency-controlled requests (works everywhere)
• native – provider batch APIs (OpenAI, Gemini)
• auto – native if available, otherwise emulated

Results

Always mapped back by item_id.

{
  "a": ResultOk(...),
  "b": ResultErr(...)
}

Partial success is normal and expected.

---

Providers

Currently supported:

• OpenAI
• Gemini
• Local echo (testing)

Providers are internal adapters. You do not need to build a gateway.

---

Persistence

Each job is stored in SQLite:

• survives process restarts
• supports retries
• allows progress inspection

job.progress()
job.status()

---

Logging

Enable lightweight progress logs:

export OPENAPI_BATCH_LOG=1

You’ll see submission, polling, heartbeats, and completion.

---

Testing

Unit tests (fast)

• local_echo provider
• no network

Integration tests (real APIs)

export OPENAI_API_KEY=...
export GEMINI_API_KEY=...

pytest -m integration

Integration tests are opt-in and may incur cost.

---

What this library does not try to do

• It does not abstract prompts
• It does not hide provider semantics
• It does not invent new APIs

It focuses only on batch execution, reliability, and DX.

---

License

MIT