OaiBatch (Swift)

A macOS app for OpenAI's Batch API (Responses API). Create batch requests, monitor status, and fetch/copy responses when they complete (up to 24h turnaround).

Warning: Use at Your Own Risk

This software interacts with OpenAI's Batch API, which may have bugs that can result in unexpected charges.

I experienced an issue where a single batch job triggered multiple executions on OpenAI's side. The batch listing showed only one job, but the logs revealed multiple complete executions (each with full input and successful output). I was charged for all of them, and the batch never completed. This appears to be an OpenAI bug, not an issue with this tool.

• I contacted OpenAI about the above and I noticed that on the next day I was billed the right amount, so apparently they have a cron or something fixing the batch cost overnight... so you might expect to see a single batch job listed, lot of requests on the logs, usage will show a higher than expected billing cost, it should be fixed overnight and everything will be back to normal. I am still keeping a close eye on it.

Recommendations:

• Monitor your OpenAI usage dashboard while jobs are running
• Set up billing alerts on your OpenAI account
• Be cautious with expensive models (gpt-5.2-pro, o3-pro) and high token limits
• Check logs for duplicate executions if a batch takes unusually long
• Start with smaller/cheaper requests to verify everything works as expected

This tool is provided as-is with no guarantees. You are responsible for any charges incurred.

Screenshots

Requirements

• macOS 14+
• Xcode 15+ (to build from source)
• OpenAI API key via OPENAI_API_KEY or ~/.oaibatch/config.json

Build & Run

Xcode

Open OaiBatch-Swift/OaiBatch.xcodeproj and run the OaiBatch scheme.

Scripted build (app + DMG)

cd OaiBatch-Swift
./build.sh all

Outputs:

• OaiBatch-Swift/build/export/OaiBatch.app
• OaiBatch-Swift/OaiBatch.dmg

Usage

1. Launch the app.
2. In Settings, save your API key (stored at ~/.oaibatch/config.json). OPENAI_API_KEY overrides the saved key.
3. Create a request (model, reasoning effort, max output tokens, optional web search).
4. Refresh statuses and fetch/copy responses when completed.

How it works

1. Create: Uploads a JSONL file with your request to OpenAI and creates a batch job
2. Requests: Refreshes status of your batches from the API
3. Response: Downloads the output file and extracts your request's response text

Requests are stored locally in ~/.oaibatch/requests.json for tracking.

Configuration

• Model: Selectable per request (default: gpt-5.2-pro)
• Reasoning effort: none, low, medium, high, xhigh (none disables reasoning)
• Web search: Optional web_search tool (context size: low / medium / high)
• Endpoint: /v1/responses
• Completion window: 24 hours

Batch pricing (per 1M tokens)

Model	Input / 1M	Output / 1M
gpt-5.2	$0.875	$7.00
o4-mini	$0.55	$2.20
o3	$5.00	$20.00
o3-pro	$10.00	$40.00
gpt-5.2-pro	$10.50	$84.00

Cost estimates shown in the app use the model stored on each request.

License

MIT