|
| 1 | +--- |
| 2 | +title: "AI Collection Creator" |
| 3 | +description: "Design and create multiple MongoDB collections at once by chatting with an AI agent inside the urBackend dashboard." |
| 4 | +--- |
| 5 | + |
| 6 | +The AI Collection Creator is an in-dashboard agent that turns a natural-language description of your app into a set of production-ready MongoDB collections. You describe what you're building, iterate on the schema through chat, review the design on an interactive canvas, and create every collection in your project with a single click. |
| 7 | + |
| 8 | +Use it when you want to bootstrap a new project's data model quickly instead of clicking through the manual field builder one collection at a time. |
| 9 | + |
| 10 | +## When to use it |
| 11 | + |
| 12 | +- **New project bootstrap** — turn a product idea ("multi-tenant SaaS with workspaces, projects, and tasks") into 5–15 collections without hand-designing each one. |
| 13 | +- **Exploring a schema shape** — iterate on relationships and field types in chat before committing anything to the database. |
| 14 | +- **Reference designs** — ask for well-known patterns (e-commerce, CMS, subscription billing) as a starting point you can then edit manually. |
| 15 | + |
| 16 | +If you already know the exact shape of a single collection, the manual builder is faster. See [Collections & Schemas](/concepts/collections-schemas) for the manual flow and field-type rules. |
| 17 | + |
| 18 | +## Open the AI Collection Creator |
| 19 | + |
| 20 | +1. In the dashboard, open your project. |
| 21 | +2. Go to **Database** and click **Create Collection**. |
| 22 | +3. At the top of the create-collection page, switch the mode selector from **Manual Builder** to **AI-Assisted**. |
| 23 | + |
| 24 | +The page splits into two panels: |
| 25 | + |
| 26 | +- **Left — AI Schema Assistant.** A chat with the agent, plus a model selector and quick-start prompts. |
| 27 | +- **Right — Schema Canvas.** An interactive visualizer that renders the proposed schema as connected cards, one per collection. Fields, types, required/unique flags, and `Ref` relationships are drawn as edges between cards. |
| 28 | + |
| 29 | +## Chat with the agent |
| 30 | + |
| 31 | +Type a description of your app into the composer and press **Enter** (or **Shift+Enter** for a newline). The agent responds with either: |
| 32 | + |
| 33 | +- A clarifying question, if it needs more detail before designing a schema. |
| 34 | +- A proposed set of collections, which appear as cards on the Schema Canvas on the right. |
| 35 | + |
| 36 | +You can keep chatting to refine the design — add fields, split a collection, change a type, add references, remove something you don't need. Each turn updates the Schema Canvas in place. |
| 37 | + |
| 38 | +### Quick-start prompts |
| 39 | + |
| 40 | +When the chat is empty, the assistant offers a few one-click prompts you can send as-is or edit. They currently cover: |
| 41 | + |
| 42 | +- **E-commerce Platform** — products, categories, orders, customers, product reviews. |
| 43 | +- **Project Management SaaS** — workspaces, projects, tasks, milestones, team members. |
| 44 | +- **Subscription Billing API** — plans, subscriptions, invoices, payment methods, usage logs. |
| 45 | +- **Content Management (CMS)** — articles, authors, categories, media assets, publishing revisions. |
| 46 | + |
| 47 | +Use these as starting points, then iterate in chat. |
| 48 | + |
| 49 | +### Model selector |
| 50 | + |
| 51 | +A dropdown at the top of the chat panel lets you pick the model used for the next turn. The default (`qwen/qwen3.6-27b`) and `openai/gpt-oss-120b` currently produce the best schema designs; smaller models are faster but may return simpler results. |
| 52 | + |
| 53 | +Individual messages are limited to 2,000 characters. Longer inputs return a `400` error asking you to shorten the message. |
| 54 | + |
| 55 | +## Review the schema on the Schema Canvas |
| 56 | + |
| 57 | +Every proposed collection is rendered as a card on the canvas on the right: |
| 58 | + |
| 59 | +- Each row is one field with its type. |
| 60 | +- **Required** and **unique** flags are marked on the field. |
| 61 | +- `Ref` fields draw an edge to the referenced collection so you can see relationships at a glance. |
| 62 | +- `Object` and `Array` fields show their nested shape. |
| 63 | + |
| 64 | +You can pan and zoom the canvas to inspect large designs. The canvas updates whenever the agent returns a new schema in chat. |
| 65 | + |
| 66 | +## Create the collections |
| 67 | + |
| 68 | +When you're happy with the design, click **Insert All** above the Schema Canvas. urBackend asks you to confirm, then creates every collection in the design in a single bulk operation. When the call finishes: |
| 69 | + |
| 70 | +- Each collection is either marked as created or listed with the reason it failed (for example, a name that already exists in the project). |
| 71 | +- A toast summarizes how many collections were created and how many failed. |
| 72 | +- The chat session is cleared so you can start a fresh design. |
| 73 | + |
| 74 | +### Bulk creation limits |
| 75 | + |
| 76 | +The bulk-create endpoint enforces the following: |
| 77 | + |
| 78 | +- Maximum **20 collections per Insert All call**. Designs larger than 20 collections must be split across multiple sessions. |
| 79 | +- Your project's overall collection limit (from your plan) still applies. If Insert All would exceed it, the excess collections fail with `Collection limit reached`. |
| 80 | +- Duplicate names against collections that already exist in the project fail with `Collection already exists`. Other collections in the same batch still succeed. |
| 81 | + |
| 82 | +### The `users` collection is reserved |
| 83 | + |
| 84 | +The AI Collection Creator never creates a collection named `users`. It's stripped from the AI response before the schema reaches the canvas, and the bulk-create endpoint additionally rejects any collection named `users` with a `403`. |
| 85 | + |
| 86 | +The `users` collection has a fixed schema contract used by the auth system. Create it manually from the **Manual Builder** tab, where the dashboard pre-fills the required `email` and `password` fields. See [Collections & Schemas — The users collection](/concepts/collections-schemas#the-users-collection) and the [Authentication guide](/guides/authentication). |
| 87 | + |
| 88 | +## Session iteration limits |
| 89 | + |
| 90 | +A single AI Collection Creator session is capped at **20 iterations** (user + assistant turn pairs) when using the urBackend platform key. When you approach the limit, the composer shows a counter (`N of 20 AI turns used`). When you hit it, the composer is disabled and the next request returns: |
| 91 | + |
| 92 | +``` |
| 93 | +403 Forbidden |
| 94 | +Session iteration limit reached (20/20). Add your Groq API key in Settings for unlimited usage. |
| 95 | +``` |
| 96 | + |
| 97 | +Chat transcripts are cached for 2 hours. Clicking **Reset chat** in the canvas panel clears the transcript, resets the iteration counter, and starts a fresh session. |
| 98 | + |
| 99 | +### BYOK bypasses the iteration cap |
| 100 | + |
| 101 | +If you've configured a personal Groq API key at the developer or project level, the iteration cap **does not apply** — you can iterate as many turns as you like in a single session, subject only to Groq's own rate limits and billing on your key. |
| 102 | + |
| 103 | +BYOK also removes the platform-key monthly AI session cap that applies to other AI features. See [BYOK for AI](/guides/ai-byok) for how to configure a key. |
| 104 | + |
| 105 | +## Related |
| 106 | + |
| 107 | +- [Collections & Schemas](/concepts/collections-schemas) — field types, constraints, and the manual builder. |
| 108 | +- [BYOK for AI](/guides/ai-byok) — bring your own Groq key to bypass iteration and session caps. |
| 109 | +- [Authentication](/guides/authentication) — how the reserved `users` collection is used. |
0 commit comments