> For the complete documentation index, see [llms.txt](https://help.felt.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.felt.com/felt-ai/ai-assistant.md). # AI assistant {% embed url="" %} The AI assistant is built into every Felt map. Open the chat and ask for what you want — a dataset to find, an analysis to run, a style to apply — and it gets to work. Review the results, refine them in conversation, and share when you're done. {% hint style="success" %} Felt AI is included in the [Enterprise plan](https://felt.com/pricing) and available as an optional add-on on the Professional plan. {% endhint %} ## What can the assistant do?
Ask for what you want and the assistant handles the rest — here's what that looks like in practice. **Find and add data.** The assistant finds the data you ask for, starting with Felt's Global Library of curated public datasets and everything connected to your workspace — your maps' layers, the workspace library, Felt Servers, and cloud databases. Need something more? It can search public data catalogs too. > Find flood zones for Harris County and add them to this map. **Query your databases.** Connected to Snowflake, BigQuery, Databricks, Postgres, Redshift, or SQL Server? Ask in plain language and the assistant writes the SQL, runs it, and turns the results into layers. > Which of our properties are within 5 miles of a wildfire hazard zone? Use the portfolio table in Databricks. **Run spatial analysis.** Buffers, joins, filters, aggregations, and more — the assistant creates new layers from your existing ones. > Count the number of permits in each neighborhood and add the result as a new layer. **Style your map.** The assistant styles layers based on what's actually in your data — categorical palettes, numeric ramps, heatmaps, H3 hexbins, labels, and popups — using [Felt Style Language](https://developers.felt.com/felt-style-language) under the hood. > Style this as a choropleth by median income, with a legend that reads well on the light basemap. It also handles the smaller things: zooming to what matters, showing or hiding layers and legend items, pulling in data from URLs (ESRI services, WMS, GeoJSON, and more), and naming your map. As it works, the chat shows each step — what it's reading, what it's running, what changed on the map — so you can follow along and jump in at any point. ## Opening the assistant There are two ways to open the assistant on any map: * Click the **Felt AI** button (![](/files/VFQbTbSJKlj2w3U1n6D1)) in the toolbar * Press **Cmd+Shift+K** (Mac) or **Ctrl+Shift+K** (Windows) The chat panel opens on the right side of your map. If you're not sure where to start, try one of the suggested prompts — or just ask the assistant what it can see and do. ## Tour the chat window * **Conversation dropdown** — shows the current conversation's title. Click it to switch between your conversations on this map. * **New chat** — starts a fresh conversation. Use this when you're switching to an unrelated task; a focused conversation gets better results. * **Close** — closes the panel. Your conversation is saved and will be right where you left it when you reopen the assistant. * **Prompt box** — type your request and press Enter. Type **@** to reference a specific layer or data source (see below). * **Microphone** — speak your prompt instead of typing it. Dictation adds to whatever you've already typed; click the microphone again or press **Esc** to stop. Available in browsers that support speech recognition. * **Response actions** — hover over any response to **copy** it or give **feedback**. ## Edit mode vs. view mode **In edit mode** — the assistant works on the live map: what it adds, styles, and computes appears immediately for you and your collaborators. **In view mode** — submitting a prompt creates **your own duplicate of the map**, opened in edit mode with your prompt carried over. The original is never modified. Your duplicate is saved to your workspace and shows a card linking back to the source map. (Have Edit access? You can choose to edit the map directly instead.) {% hint style="info" %} Using the assistant in edit mode requires a Full Access seat with Edit access on the map. Anyone in the workspace can prompt from view mode, since the work happens on their own copy. {% endhint %} ## Conversations You can have multiple conversations with the assistant on every map. They're saved with the map and titled automatically as you work: * **Continue where you left off** — reopening the assistant loads your most recent conversation; it remembers what you discussed and built. * **Start fresh** with **New chat** when you change topics. When you have previous conversations on a map, the panel's start screen lists your recent chats so you can jump back in. * **Switch between conversations** from the dropdown at the top of the panel — useful when you're iterating on styling in one thread and data questions in another. ## What the assistant can see The assistant sees what you can see — nothing more. It can read: * **The current map** — layer names, schemas, and (for queryable layers) the data itself * **Your workspace** — the library, Felt Servers, and connected data sources * **Felt's Global Library** — curated public datasets * **Public GIS sources** — when you ask it to search for data Queries against your databases run with your existing connection's access, and Felt never trains models on your data. Type **@** in the prompt box to point the assistant at exactly the data you want. You can reference: * Layers on the current map * Layers in your workspace library * Layers in Felt's Global Library * Connected data sources (your databases) > Compare @Census Tracts with @Store Locations and highlight tracts with no store within 2 miles. You don't have to use **@** — but referencing removes ambiguity when names are similar or your workspace contains many sources. ## Sharing what you build Everything the assistant produces is a normal part of your Felt map — layers, styles, and analysis results. Share the map the way you always do (the **Share** button), and viewers see the results without needing Felt AI access themselves. ## Felt AI beyond the assistant This page covers the AI assistant — the chat panel on the map, where Felt AI has its full toolkit. Felt AI also shows up in other places, scoped to the job at hand: * In the [Extensions](/developer-platform/extensions.md), [custom popups](/developer-platform/custom-popups.md), and [SQL queries](/data-sources/cloud-sources/sql-queries.md) editors, Felt AI drafts that editor's code, which you can read and edit. See [More ways to work with AI](/felt-ai/more-ways-to-work-with-ai.md). * Through [MCP integrations](/felt-ai/mcp.md), external AI tools like Claude and ChatGPT get Felt AI's capabilities in their own interfaces, with your workspace permissions. ## Frequently asked questions
Who can use Felt AI? Felt AI is included in the Enterprise plan and available as an add-on on the Professional plan (not on legacy plans). Working in edit mode takes a Full Access seat with Edit access on the map. Anyone in the workspace can prompt from view mode and gets their own copy of the map.
Does the assistant change my map without asking? In edit mode, yes — it applies work as it goes, narrating each step so you can see what changed and ask it to adjust or undo. In view mode it only ever works on your own duplicate.
Can I see or edit the SQL and styling the assistant generates? No. The assistant delivers results — its SQL runs behind the scenes and its styling lands directly on the layer, where you can adjust it in the style editor afterwards. For AI-drafted code you can edit, use [SQL queries](/data-sources/cloud-sources/sql-queries.md), [custom popups](/developer-platform/custom-popups.md), or [Extensions](/developer-platform/extensions.md).
What data can the assistant access? Only what you can access: the current map, your workspace library and connected sources, the Global Library, and public sources when you ask it to search. Felt never trains on your data.
Are my conversations private? They're saved with the map so you can pick up where you left off. The changes the assistant makes in edit mode are visible to everyone with access to the map, like any other edit.
Is Felt AI secure? Yes. Database connections are read-only unless you grant write access, and SQL queries only ever run `SELECT` statements. Felt AI works within your existing SSO and permissions, and its actions are logged like any other activity in your workspace. Queries run live against your connected databases, in their native SQL, rather than copying your data elsewhere. Felt doesn't train models on your data, and none of it becomes public. Felt AI runs on frontier models hosted on Amazon Bedrock.
--- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://help.felt.com/felt-ai/ai-assistant.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.