The AI Assistant calls an AI provider's model with an API key you registered, and the cost is charged to that key. In the menu you register keys, set the response language and view usage. AI and MCP explains where keys are stored and what data goes to the AI provider.
Open My page by clicking your user display at the right of the header and then the My Page item. Expanding the
menu shows the screen, the screen and the screen. These three screens are not locked even without signing in. The Add AI model button below the key list of the AI Assistant panel also opens the screen, and if you have no key yet, the registration window opens right away.What's on the screen
| No. | Name | Description |
|---|---|---|
| 1 | menu | Expand it to see the three screens under it. |
| 2 | Account API Keys group | Keys stored encrypted in your account. Shown only when you are signed in. The same keys are used on your other devices and on the web. |
| 3 | Key card | The provider, name, primary key mark (★), the key with only its last four characters shown, and the Edit and Delete buttons. Clicking the empty star (☆) makes that key the primary key of its group. Keys turned off carry a Disabled label. |
| 4 | Settings | The model this key calls, the upper limit of a single response's length (Max output tokens), the Monthly budget item (Account API Keys group only) and any address you set yourself. |
| 5 | Usage in the last 30 days | The key's call count, estimated cost and input and output tokens. If a key in the Account API Keys group has a monthly budget, a bar shows this month's cost against it. |
| 6 | API Keys on This Device group | Keys stored only on this device. Shown on the desktop only. The raw key is kept in the operating system's secure storage. |
| 7 | Add Key button | Opens the window for registering a key in that group. |
Which groups you see, and which keys you can pick in the AI Assistant, depend on the environment. The project type does not matter.
| Environment | Groups shown | Keys you can pick in the AI Assistant |
|---|---|---|
| Web | Account API Keys group | The Account API Keys group, except keys whose address is http, localhost or a private IP |
| Desktop, not signed in | API Keys on This Device | API Keys on This Device |
| Desktop, signed in | The Account API Keys group and the API Keys on This Device group | Both |
- Each group has one primary key. The AI Assistant uses the Account API Keys group's primary key first, and the API Keys on This Device group's primary key if there is none.
- Keys turned off do not appear in the AI Assistant's key list.
Registering an API key
- Click the Add Key button (7) of the group that should hold the key. If the group is empty, click the Add First Provider button in the middle.
- Pick the provider and paste the API key. The link below the key field opens that provider's key page.
- Pick the model in the Model field. The list holds that provider's models. To also see fixed versions with a date, turn on the snapshot checkbox.
- Fill in the Label (optional) field, the Max output tokens field and the Monthly budget field (Account API Keys group only) if you need them.
- Check that the key and address work with the Test connection button, then click the Save button.
| No. | Name | Description |
|---|---|---|
| 1 | Enabled switch | When off, AI is never called with this key. The primary key cannot be turned off. |
| 2 | Set as primary key switch | When on, this becomes the primary key of the group. The first key of a group becomes the primary key automatically. |
| 3 | Provider field | Keys in the API Keys on This Device group choose from OpenAI, Claude (Anthropic), Gemini (Google) and Ollama. Keys in the Account API Keys group choose from the list the NeoSQL server provides. It cannot be changed after registration. |
| 4 | Private endpoint field | Ollama needs a server address. For other providers, fill it in only when going through somewhere other than the default address, such as an in-house gateway or proxy. A desktop-only address gets a notice next to the field. |
| 5 | Model field | The model the AI Assistant calls with this key. Keys in the API Keys on This Device group and Ollama also accept a model name typed in that is not in the list. |
| 6 | Max output tokens field | The upper limit of a single response's length. Too small and SQL gets cut off; too large and cost and time grow. The help next to the field recommends 4096. |
- When editing, leaving the API Key field empty keeps the saved key. Typing a new value replaces it.
- The primary key cannot be deleted. Make another key the primary key first. Past usage of a deleted key is kept.
- The Monthly budget item is in US dollars; leave it empty for no limit. Once the month's estimated cost exceeds the budget, AI cannot be called with that key. Keys in the API Keys on This Device group have no budget.
Using models you run on this device or in-house
Models you run on your PC or an in-house server, such as Ollama, are also registered as keys. For Ollama, choose Self-hosted or Ollama Cloud in the Ollama connection field.
- Pick Ollama in the Provider field and leave the Ollama connection field as Self-hosted.
- Type the server address in the Private endpoint field. Ollama on your PC is
http://localhost:11434. - Click the Load models button to read the server's model list, and pick one in the Model field. Expanding the Model details area shows the Processing location item, the Size item and the Maximum context (tokens) item.
- If the server requires authentication, turn on the Use server authentication checkbox, enter the access key and save.
AI requests leave from this device on desktop and from the NeoSQL server on the web. Choose the environment for the AI server address you use.
| Address | Where to use it |
|---|---|
Public https address | Web and desktop |
http address | Desktop |
localhost, private IPs (10.x, 192.168.x and so on) | The desktop app on a device that can reach the address |
- You can register such addresses in the Account API Keys group too. The registration window shows the desktop-only notice, and the key is used for model lookup and AI calls only on the desktop.
- A self-hosted Ollama can also call Cloud models. The address alone does not guarantee that processing happens on your PC, so check the Processing location item in the Model details area.
- Use models that support tool calling with AI Assistant. Check support in the Model details area.
- Check Ollama usage through token and call counts.
Response language and common instructions
On the screen you set what applies to every AI response.
| Item | Description |
|---|---|
| Response language field | With Auto, the AI answers in the language of your question. Pick a language and it answers in that language whatever the question's language. SQL and identifiers are not changed. |
| Response instructions field | Instructions added to every response. They follow NeoSQL's own instructions and cannot change the security policy, execution limits or response language. |
- Click the Save button and the settings apply from the next request.
- They are saved in your account when you are signed in, and on this device when you are not.
Viewing usage
The screen collects AI calls from the last 30 days.
| No. | Name | Description |
|---|---|---|
| 1 | Account API Keys group usage | The Account API Keys group's Active keys item, the 30-day calls item, the 30-day cost (estimated) item and the Estimated this month item. If every key has a monthly budget, a bar shows usage against their sum. Shown only when you are signed in. |
| 2 | Per-feature table | The features are Text to SQL, Query Tuning, Sample Data generation and Ask about SQL error. Per feature, the Calls, Input, Cache read, Output, Reasoning and Cost (estimated) columns. The Cache read and Reasoning columns are separate values not included in the Input and Output columns. |
| 3 | API Keys on This Device usage | The 30-day calls item and tokens made with the keys in the API Keys on This Device group, per model. Shown on the desktop only. Costs are not calculated. |
- Usage is counted from the requests and responses NeoSQL recorded. Costs are estimates, so check the actual charges in the provider's console.
- When an external AI calls NeoSQL through MCP, NeoSQL does not call a model, so nothing is recorded here.
Things to keep in mind
- Keys in the Account API Keys group are stored encrypted on the NeoSQL server and are used on the web and on every device where you sign in. They stay in your account after you sign out.
- Keys in the API Keys on This Device group exist only in this device's operating system secure storage. They do not move to other devices and cannot be used on the web.
- Where the operating system's secure storage is unavailable (such as Linux without a keychain), keys in the API Keys on This Device group are stored in plain text. A The API key will not be encrypted confirmation appears before saving.
- The Monthly budget item is judged by the cost NeoSQL estimates. It may differ from the provider's actual charges, so set a limit on the provider side too.
- A key's group cannot be changed later. To move a key from the Account API Keys group to the API Keys on This Device group, register it again in the API Keys on This Device group.
Related screens
- AI Assistant: the panel where you pick a registered key and chat.
- AI and MCP: the difference between the Account API Keys group and the API Keys on This Device group, and the data that goes to the AI provider.
- Project AI settings: the window that allows AI use per connection.
- Security and where your data goes: where keys are stored and the traffic that leaves your device.