What Is Flexnote AI? A Knowledge Agent for Notes and Whiteboards (2026)

As a library grows, two problems often appear: notes you know you wrote are hard to find when keywords do not match, and ideas on a whiteboard stay scattered without becoming structure. Many people paste a fragment into ChatGPT. That helps for a moment, but the knowledge base and the chat remain separate systems.

Flexnote AI connects them. Notes and boards stay local. You can query your library with your own model key — answers include citations — and let the Agent edit boards, create cards, and manage todos. It is available to Pro members, and upgrading gifts smart-index points for semantic search and image OCR. Settings show three parts: Knowledge Chat, card AI Actions, and smart indexing.

1. What Pro members get

Flexnote AI is a Pro benefit covering knowledge-base Q&A, board editing, and card Actions. Models run on your own API key: choose the provider, model, and Base URL. DeepSeek, Tongyi, and others are supported; you settle with the provider you choose and are not locked to a single vendor.

Upgrading also gifts 3,000 smart-index points for semantic search and image OCR. At about 500 characters per ordinary note, that covers roughly 4,000 notes. When you need more, add Lite / Standard / Heavy packs; purchased points remain valid long-term. Details are on the pricing page.

2. Setup in a couple of minutes

Open Settings → AI assistant (there is also an entry in the Chat title bar). Enable AI, choose a provider, fill in Key / Base URL / model, then run Test connection.

Built-in providers include OpenAI, DeepSeek, Tongyi Qianwen, Moonshot/Kimi, SiliconFlow, and Zhipu GLM, plus custom OpenAI-compatible endpoints. Chinese models can be selected directly; you do not need to route everything through OpenAI.

3. Ask and Agent: query vs. execution

Open Knowledge Chat from the top bar or a detail-page AI icon. The two modes have different roles.

Ask focuses on lookup and synthesis. You can @-mention a card for context. Answers include citations, and may show reasoning and tool steps. Save a useful reply as a card. For example: "Help me synthesize recent notes on XX" or "What claim do these cards share?"

Agent focuses on execution. Ask stays mostly read-only; Agent can write to the library and edit the board. For example:

• "Turn recent notes on XX into a whiteboard structure"
• "Tag this batch of cards and create a few todos"
• "Add sticky notes, links, and groups on this board"

Many tools are available; you do not need to memorize them. Library search, card read/write, journals and todos, whiteboard nodes and layout, video frame capture and highlights are invoked as needed. When confirmation is required, the Agent pauses and waits. The short demos below cover common scenarios.

4. Four common uses

1. Ask about video content

After importing a video, you can ask about its content. Answers include a timestamp and the matching frame, so you do not need to scrub the timeline manually. Once you find the place, decide whether to capture a frame, highlight it, or save a timestamp card.

For annotation features themselves, see best video annotation tools.

2. Organize key points while reading papers

Import a paper or report, distill key points into cards, and read them beside the source. This is especially useful early in a literature review: lay out claims and evidence trails, then return to PDF annotation to verify and rewrite.

3. Generate a mind map in chat and save it to the library

A mind map can be generated during a conversation — for example while summarizing a PDF. After you are satisfied with the preview, add it to the card library. From there you can edit it again and reuse it on other boards or topics, instead of leaving it only in chat history.

4. Fill tags and properties in bulk

After a library has grown for a while, the more common issue is missing classification and empty property columns, not writing one more card. The Agent can create tags and property columns, then fill values card by card so views can filter later. You can also bring cards under a tag into context for a summary and relationship map.

5. Quick actions on the card menu

Click a card on the whiteboard to open AI Actions: summarize, generate a mind map, translate (ZH↔EN), or a custom action (name and prompt, with optional pin). The current card becomes the context for an Agent session that runs the prompt. For single-card tasks, you do not need a long conversation.

6. Smart indexing: gifted points and top-ups

Once AI is enabled, smart indexing runs automatically. It helps the app understand note text and text inside images, supporting semantic search, Q&A, and content recall. Locally you still have SQLite full-text plus vectors. Cloud Embedding and OCR are metered in points. Purchased points remain valid long-term and do not reset monthly.

Top-up packs are priced low. Rough estimate for ordinary notes of about 500 Chinese characters each:

For a personal library, the Standard pack often lasts a long time. Check index status and remaining points in Settings; from the Chat sidebar you can also start an index job for the current context or force a rebuild. Even if points run out temporarily, keyword search still works, and purchasing more is straightforward.

7. Compared with a bolted-on AI page

In some note apps, AI behaves like a separate chat window: you ask a question and receive Markdown, while the library itself barely changes. In Flexnote, answers include sources you can open, and the Agent can edit the board directly. You choose the model; data stays local by default. PDF, video highlights, and OCR feed the same knowledge path. Local full-text search remains available even when index points run out temporarily.

For cards and canvas: What is Flexnote. Against Notion: Flexnote vs Notion.

8. Is it a fit?

9. Suggested trial steps

1. Configure DeepSeek or Tongyi Qianwen and confirm Test connection passes.
2. Use Ask: "Help me synthesize my recent notes." Check whether citations are accurate.
3. @ a long card and run Actions for summarize or mind map.
4. Use Agent: "Group related cards on this board by topic and connect them."
5. Save a useful answer as a card and confirm it was written to the library.

After upgrading to Pro and configuring a key, the steps above are enough for a first trial. For the core product, see the full Flexnote guide; plans and pricing are on the pricing page.

10. Closing

Written explanations only go so far. The better test is to try it: take a paper you are already reading, or a board that already has content, configure a key, ask a few questions, make a few edits, and check whether citations are accurate and results are written back to the library. In a few minutes you can tell whether it fits your workflow.