# Start a labeling and annotation queue

Items to label, who has them, the labels given and where annotators disagree. Each annotator sees only their batch, and nobody sees another person’s label first.

## An example

**You:** Load the 2,000 support replies from this CSV for labeling: helpful, unhelpful or unsafe. Two annotators per item. Give Lena and Omar the first 500.

**ADoor:** Added 2,000 entries to Items and 1,000 to Assignments

**Labels**

| Item | Annotator | Label | At |
| --- | --- | --- | --- |
| reply-0007 | Lena | Helpful | Oct 20 10:02 |
| reply-0007 | Omar | Unhelpful | Oct 20 10:15 |
| reply-0008 | Lena | Helpful | Oct 20 10:03 |

## What you can ask later

- “How far along is each annotator?”
- “Where do Lena and Omar disagree?”
- “What’s the agreement rate per label?”
- “Export the labeled items as JSON for training.”

## Who can see what

- **Each annotator** sees only their assigned items and adds labels, without seeing other people’s labels first.
- **The project lead** sees everything and resolves disagreements.
- **A model** can pre-label with add-only access to a separate field.
- **An external labeling vendor** is limited to one batch, and loses access on the next request when the work is done.

## How it compares with Label Studio or Scale

Label Studio and vendors like Scale are better at labeling interfaces for images, audio and spans, with workforce management.

For text and simple labels where access matters, ADoor is enough: each annotator sees only their batch, labels stay hidden from each other, and plain SQL measures agreement.

## How to start

1. Sign in to ADoor and connect an assistant like ChatGPT. Claude, Grok and Cursor connect the same way, through MCP.
2. Tell it: “Set up a labeling queue: Items (id, text, batch), Assignments (item, annotator) and Labels (item, annotator, label, note, at).”
3. Invite annotators with access to their own assignments (one grant per person), and give a labeling vendor one batch.

[Get your account](https://app.adoor.halmoslabs.com/signup)

## Questions people ask

### Can annotators label from chat?

Yes, through their assistant, or with a small interface on the HTTP API.

### Can I export for training?

Yes, as CSV or JSON.

### Do I have to pay?

Not on the free plan, which holds 3 tables and 100 rows. Pro is $10 a month per workspace, with 1 GB of storage and no per-GB charges.

### Which assistants can I use?

Any of ChatGPT, Claude, Grok or Cursor, connected through MCP. Each one asks for your approval, and you decide which lists it can use.

## Related

- [Start a dataset registry](/use/dataset-registry/)
- [Track eval results over time](/use/eval-results/)
- [Run a task queue for your agents](/use/agent-task-queue/)
- [ADoor vs Airtable](/compare/adoor-vs-airtable/)
- [ADoor vs Supabase](/compare/adoor-vs-supabase/)
- [All use cases](/use/)

Updated October 20, 2026.