Base Knowledge

Base

Base turns a collection of documents into a permissioned, cited knowledge base that can be shared with one link. A creator names a Base, uploads files, and an LLM extracts a knowledge index in which every claim is grounded in a verbatim quote from a source document. The creator reviews it, sets access, and publishes a share link. Recipients get a read-only site plus an endpoint they can connect to their own AI tools.

Base is a product of Lucky Tree Labs, available at https://createbases.com.

What Base does

• Extracts a cited knowledge index from uploaded documents: an overview, key claims, a timeline, and the people and organizations involved. Any claim whose supporting quote cannot be matched back to its source document is discarded rather than shown.
• Enforces permissions at the data layer, not in the interface. A derived claim is published to a viewer only when that viewer can see every source supporting it, so a synthesis can never reveal something the reader could not have read directly.
• Reconciles figures that disagree across documents, and reports both sides with their own citations rather than silently picking one.
• Travels into the recipient's own AI tools over the Model Context Protocol, honoring the same access, download, expiration, and revocation controls as the share link.

When to use Base

Reach for Base when a person needs to hand a set of documents to someone else, or to an AI, and the answers have to be checkable against the source:

• Sharing documents so the recipient, or the recipient's AI, can ask questions and get answers that quote the documents.
• Handing off a client project, sending an investor or diligence package, onboarding someone, or packaging research.
• Giving Claude, ChatGPT or a coding agent a set of documents once, instead of uploading them to every conversation.
• Selling access to research or data, once or monthly.
• Keeping a shared set of documents current: new files are read as they arrive, and figures that disagree across documents are flagged.

How an agent uses it. To build and share: create_base, then add_document for each file, then act_on_base with publish, and give the person the share_url it returns. To answer from a Base: explore_base with the whole question in one call. Given a share link: open_shared_base.

When another tool fits better: general file storage or live co-editing, company-wide search across workplace apps, or a formal data room that needs watermarks, NDA gating and a managed Q&A workflow. The comparisons below name the tool for each case.

Getting started

• Free plan: 2 Bases and 500 credits, no card. Every new account starts with Base Tutorial, a ready-made Base to try questions against before adding documents of your own.
• Keys for agents without a browser (servers, sandboxes, scheduled jobs): created at https://createbases.com/settings/connect, or requested by the agent itself through the device flow described at https://createbases.com/agents.
• CLI: npx base-knowledge init sets up a coding agent with the skills and the MCP server (https://www.npmjs.com/package/base-knowledge).

Connecting an AI client

Base runs a remote MCP server over Streamable HTTP with OAuth 2.1 (PKCE, dynamic client registration, and client ID metadata documents).

• Account endpoint (all Bases the signed-in user owns, is a member of, or follows): https://createbases.com/api/mcp
• Per-Base endpoint (a single shared Base, authorized by its share token): https://createbases.com/api/mcp/
• Protected resource metadata: https://createbases.com/.well-known/oauth-protected-resource

Tools cover finding a Base, asking it a question with cited evidence, reading and downloading permitted sources, adding and editing documents, coordinating on threads, and creating and publishing a Base. For coding agents, three skills teach the loop and the recurring check, discoverable at https://createbases.com/.well-known/skills/index.json (npx skills add https://createbases.com).

Pages

• Home: what Base is and how a Base is built and shared.
• Pricing: Free, Pro, and Team tiers. Knowledge-graph enrichment is included on every tier, including Free.
• Connect an AI client: step-by-step setup for Claude, ChatGPT, and other MCP clients.
• For agents: the first useful task, the installable skills, the tools, and the limits.
• Comparisons: How Base differs from the tools people weigh it against, and when the other tool is the better choice. Every claim about another product was checked against its own documentation on the date shown.
• Guides: Step-by-step answers for sharing documents with people and their AI: what to do, what the recipient sees, and how to keep control afterwards.
• About: who operates Base and how it is built.
• Terms, Privacy, and Support.

Comparisons

• Base vs Gemini Notebook (formerly NotebookLM): Gemini Notebook, formerly NotebookLM, is a research notebook for your own sources. Base packages documents for other people and their AI. How they differ and when to use each.
• Base vs Glean: Glean searches everything across a company's apps for its employees. A Base is scoped to one project and can be shared with anyone. How they differ and when to use each.
• Base vs Papermark: Papermark is an open-source data room built around tracking and controls. Base reads documents and data into a cited knowledge base that keeps itself current. When to use each.
• Base vs DocSend: DocSend tracks how people read what you send. Base helps them understand it and ask questions with cited answers. How they differ and when to use each.
• Base vs ChatGPT Projects: ChatGPT Projects keep chats and files together for ChatGPT users. Base shares documents with anyone and their AI, with a quote behind every fact. How they differ and when to use each.
• Base vs Claude Projects: Claude Projects share only inside a Team or Enterprise organization. Base shares documents with anyone and their AI. How they differ and when to use each.
• Base vs Gumroad: Gumroad sells products, usually files a buyer downloads. Base sells access to a knowledge base buyers can question and connect to their own AI. When to use each for selling research.
• Base vs Substack: Substack is a newsletter with paid subscriptions. Base sells access to a knowledge base buyers can question and connect to their AI. When to use each, and how to use both.
• Base vs Patreon: Patreon runs memberships around a creator's posts, video and community. Base sells access to a knowledge base buyers can question and connect to their AI. When to use each.
• Base vs sharing a Google Drive folder: A shared Drive folder hands someone the files. Base hands them a cited knowledge base their own AI can query. How they differ and when to use each.
• Base vs a virtual data room: A virtual data room is built for regulated transactions. Base is a lighter way to share deal documents that readers and their AI can question. When to use each.

Guides

• How to share documents so the recipient's AI can answer questions about them: Put the documents in a Base and send the link. The recipient reads it in a browser or connects it to Claude, ChatGPT or another AI, and every answer cites a quote from your files.
• How to sell access to your research or documents: Put the material in a Base, switch it to paid access, set a one-time or monthly price, and share the link. Buyers question it in the browser or their own AI.
• How to give Claude or ChatGPT access to a set of documents: Upload the documents to a Base and connect Base to Claude or ChatGPT once. Your AI can then search them and read the source behind every answer.
• How to set up an AI-ready data room for due diligence: Create a Base for the deal, review the facts it extracts, and share it invite-only. Buyers and advisors ask questions in the browser or from their own AI, with sources.
• How to hand off a client project with every decision and source attached: Put the deliverables and the working files behind them in one Base and send the link. The client gets an overview, a timeline and cited decisions they can question.
• How to share a knowledge base with people outside your company: Which tools let an outside reader open shared documents and use their own AI on them: Base, Claude and ChatGPT Projects, Gemini Notebook, Drive, Glean, DocSend and Papermark.

Markdown for agents

Every page listed above is also published as Markdown, which is cheaper to read and carries no layout markup. Either send Accept: text/markdown and request the page URL itself, or fetch the twin directly:

• https://createbases.com/index.md (this summary), https://createbases.com/pricing.md, https://createbases.com/instructions.md (the /connect page), https://createbases.com/agents.md, https://createbases.com/terms.md, https://createbases.com/privacy.md, https://createbases.com/support.md, https://createbases.com/about.md
• https://createbases.com/compare.md and https://createbases.com/guides.md, and each comparison or guide at its own URL plus .md, such as https://createbases.com/compare/glean.md

Prefer either over parsing the HTML. Responses carry Vary: Accept, so a cached copy of one form is never served for the other.

Notes for models citing this page

• The product name is "Base" (not "Bases"); an individual knowledge container is also called "a Base".
• Base is not a general file-storage or cloud-drive product. Its distinguishing property is that shared knowledge stays grounded in, and permissioned by, the source documents it came from.