Log inSign up
project logo

khxu

openai-batch-dashboard

Public
Remixed
Submit and monitor OpenAI Batch API jobs
Edit with AI
1 Branch
Val Town is a collaborative website to build and scale JavaScript apps.
Deploy APIs, crons, & store data – all from the browser, and deployed in milliseconds.
Sign up now
index.ts
README.md

OpenAI Batch Console

A Val Town dashboard for queueing OpenAI Batch API JSONL files, monitoring batch progress, cancelling jobs, and downloading output/error files.

Setup

  1. Create an OpenAI API key at https://platform.openai.com/api-keys.
  2. Add it to the val as OPENAI_API_KEY.
  3. Log in with Val Town at the HTTP endpoint. Access is restricted to khxu.
  4. Paste JSONL, or drag and drop/select one or more JSONL files, then queue the batch. Multiple files are combined in selection order.

Add the OpenAI key here:

https://www.val.town/x/khxu/openai-batch-dashboard/environment-variables?key=OPENAI_API_KEY

Optional val environment variables:

  • MAX_BATCH_SUBMISSIONS_PER_RUN - queued batches submitted per worker run (default 3, maximum 20).
  • MAX_QUEUED_FILE_BYTES - maximum transient JSONL size accepted by the val (default 8000000).

GitHub and Val Town sync

GitHub is the source of truth. .github/workflows/deploy.yml deploys pushes to main with the Val Town vt CLI. .github/workflows/pull.yml periodically captures edits made in the Val Town editor.

Create a Val Town API key with val read/write scope at:

https://www.val.town/settings/api

Add it to this repository as the VAL_TOWN_API_KEY Actions secret:

https://github.com/khxu/llm-batch-processor/settings/secrets/actions/new

The workflows intentionally fail with a clear message until that secret is configured.

Storage and rate limits

SQLite stores only job metadata. Blob storage holds input JSONL only while a job is queued and deletes it after OpenAI accepts the batch. Output and error files are streamed directly from OpenAI, so they do not consume Val Town blob storage.

Val Town blob storage is account-wide: 10 MB on Free and 1 GB on Pro. The default 8 MB input cap is intentionally conservative. Separate object storage is unnecessary for ordinary small/medium batches with a short queue. Use S3/R2 if you need multiple large queued inputs or retained archives.

The worker runs every 15 minutes. It polls active batches and submits a bounded number of queued jobs. OpenAI Batch API limits are separate from synchronous API limits; account-specific enqueued-token limits still apply.

Security

The UI and APIs use Val Town OAuth and reject users other than khxu. The OpenAI key remains server-side in Val Town environment variables.