AI knowledge base software for grounded answers
Upload files, paste text, crawl a public site, or sync GitHub, ClickUp and Google Drive into a knowledge base. The agents you attach it to search it as they work, and every result names its document. A personal knowledge base is yours alone; an organization one opens only to the people it is shared with.
Early access. Request a demo to get started.
What a knowledge base can hold
Files, pasted text, a crawled public website, and GitHub, ClickUp and Google Drive through connectors. Bring in the documents your team already keeps; each one becomes searchable once it is indexed.
| Source | What it brings in | Limit | Updates |
|---|---|---|---|
| Files | PDF, Word and text files: TXT, MD, CSV, TSV, LOG, JSON, HTML, XML and YAML | 25 MB per file | When you upload |
| Pasted text | Text you paste in and give a name | 2 MB | When you paste |
| Public website | A public URL and the pages it links to | depth 1–10up to 100 pages | Only when you click Re-sync |
| GitHub | Issues and pull requests | Never source code | Auto-sync, about hourly |
| ClickUp | Tasks and docs | Personal knowledge bases only | Auto-sync, about hourly |
| Google Drive | Folders and files | Auto-sync, about hourly |
Auto-sync picks up changed items only.
See what is searchable
A document is searchable once it reads Indexed, and every document shows where it stands. Retire an outdated file without deleting it, and bring it back when you need it.
- Every document shows Pending, Indexing, Indexed, Failed or Deprecated
- Deprecate takes a file out of search and keeps it; Restore puts it back
- A crawled website changes only when you click Re-sync
- Connectors with Auto-sync on pick up changed items about hourly
Choose how it searches
Every knowledge base has a search mode and an embedding model. Test search shows what comes back before an agent relies on it.
- Hybrid Keyword and semantic matching together. You set the vector weight, and keyword matching takes the rest. vector 0.7 · keyword 0.3
- Vector only Semantic similarity, for questions worded differently from the document.
- Keyword only Exact terms, for names, codes and identifiers.
- Test search Preview what an agent would retrieve. Each result names its document and a relevance score.
- Hosted embeddings Fours-hosted embedding models: BGE-M3 by default, or Qwen3. BGE-M3 · Qwen3
- Your own provider Embed with a model from a provider you connect, on your own key.
Where agents can search it
An agent searches a knowledge base with a tool it calls when a question needs it, across up to three knowledge bases at a time. Every result names its document.
| Where it runs | What it can search |
|---|---|
| A custom agent | Only the knowledge bases you attach to it, each Read or Edit. Edit never deletes. |
| The built-in Insulin assistant | Your personal knowledge bases by default, read-only. |
| A channel | Up to 20 knowledge bases connected to the channel. |
| A job | The knowledge bases of the agent it runs. |
| Inbox App drafts | Your knowledge bases, read-only. |
Private until you share it
A knowledge base opens to its owner and the people it is shared with. An org admin role on its own opens nothing.
- Personal knowledge bases are visible only to you
- Org admins create organization knowledge bases
- Share with named people as Admin, Editor or User, or with everyone as Editor or User
- Fours does not use customer data to train AI models
Go deeper on knowledge bases
- Docs Knowledge Bases documentation
- Blog Which documents can an AI knowledge base read?
- Blog How to weight vector against keyword search
- Blog Keeping an AI knowledge base fresh over time
- Blog Deprecate, replace or delete a knowledge-base document
- Docs Knowledge bases in Insulin channels
Insulin is a software product from Fours — an AI platform that builds secure business apps from a description. It is unrelated to the hormone, to diabetes care, or to any medical or pharmaceutical product.
The rest of the workspace
The same knowledge bases serve the agents, channels, jobs and Inbox drafts in your workspace.
- Insulin Apps Business apps built from a description, behind roles you control.
- Agents Specialist agents for any workflow, with the model you choose per agent.
- Jobs An agent run on demand, on a schedule or when an event arrives, with every run on record.
- Agent Marketplace Ready-made agents to install and adapt.
- Inbox App Drafts replies in your voice for you to review and send.
- Pricing Pay for the AI you use, with no per-seat fees.
Frequently asked questions
What is an AI knowledge base?
An AI knowledge base is a searchable collection of your own documents that AI agents can search as they work. In Insulin, agents you attach a knowledge base to can search it, and every result names the document it came from.
What can an Insulin knowledge base hold?
PDF, Word and text files up to 25 MB each, pasted text up to 2 MB, a crawled public website of up to 100 pages, and GitHub issues and pull requests, ClickUp tasks and docs, and Google Drive folders and files.
How does hybrid search work?
Hybrid search combines keyword and semantic matching, and you set how much weight the semantic side carries. A knowledge base can also search by vector only or keyword only, and Test search shows each result's document and relevance score.
How do agents use a knowledge base?
An agent searches a knowledge base with a tool it calls when a question needs it, across up to three knowledge bases per search. Every result names its document, so an answer can show where it came from.
Which agents can see which documents?
A custom agent searches only the knowledge bases you attach to it, each Read or Edit. The built-in Insulin assistant uses your personal knowledge bases by default, read-only. A job uses its agent's, and a channel connects up to 20.
Who can open a knowledge base?
Its owner and the people it is shared with. A personal knowledge base is visible only to its owner; organization knowledge bases are created by org admins and open through a share, and an org admin role on its own opens nothing.
How does a knowledge base stay current?
Connectors with Auto-sync on sync about hourly, changed items only. A crawled website is re-crawled only when you click Re-sync. Deprecate an outdated file to take it out of search without deleting it, and restore it when you need it.
Is my data used to train AI models?
No. Fours does not use customer data to train AI models. If you connect your own model or embedding provider, that provider's terms cover what you send it.
Build a knowledge base with Insulin
See Insulin index your files, crawled sites and connectors, and what an agent finds when it searches them.