Knowledgebase
Create account

Knowledgebase Cloud by AgensTown

Knowledge you can ask, in a tree that never breaks.

Start a knowledge base from zero, let a planner propose the domain, approve it, and watch research fill every branch. You only approve or reject, so the tree stays clean.

Create account Sign in

Product tour

The real app, from an empty KB to an answer

A 90-second screen recording of the signed-in web app: create a KB, review and approve a plan, follow the research, then explore and ask the finished tree.

A real recording of the signed-in web app on a demo account, using a real KB about running and reaching AI agents. Seven chapters:

Screens

Inside Knowledgebase

Screenshots of the same app. Select one to enlarge it; use the arrow keys to move between them.

  • My KBs. Dashboard: KBs, running research and the quota meters for cloud and desktop research.
  • Tree explorer. The KB tree with a node open: definition, rules, references and its children.
  • Graph view. The same tree as a zoomable graph, centred on the selected node.
  • Research running. A research in progress: steps, 9 of 16 nodes answered and the minutes left.
  • Plan awaiting approval. The proposed tree with its questions per node, waiting for approve or reject.
  • Ask. A question ranks the nodes; the best match opens with its sourced sections.
  • Desktop and mobile apps. Downloads for the Windows desktop app and the Android app.

My KBs 1 of 7

My KBs: Dashboard: KBs, running research and the quota meters for cloud and desktop research.

Dashboard: KBs, running research and the quota meters for cloud and desktop research.

Why Knowledgebase

Structured knowledge, without the mess

Notes pile up. A strict tree does not. Every node has a place, a definition and a reason to exist.

  • Ask in plain words

    Questions are answered from the one node that owns the topic, so answers stay short, precise and traceable.

  • Plans you approve

    A planner proposes the shape of a new domain. You approve or reject it. Nothing is written before you say yes.

  • A tree that never breaks

    Classes, instances and parts, each with one parent and one definition. Every write passes the same rule gate.

  • Research that runs for you

    Our research agents fill every leaf with sourced sections, examples and pitfalls while you get on with your day.

How it works

From an empty KB to a researched domain in four steps

  1. Create a KB

    Name it. It starts with a root node and nothing else.

  2. Describe a topic

    Add a few source links if you like. Start from zero or grow under a node.

  3. Approve the plan

    Review the proposed tree with its questions. Approve or reject; that is all.

  4. Read and ask

    Every leaf gets researched. Browse the tree, open a node, ask a question.

Simulations

See it work. No account needed.

These three are scripted simulations on sample data. They run entirely in your browser: nothing is sent anywhere, no research starts and no agent runs. The real thing is one sign-up away.

Ask the KB

Simulation

Ask a question. The answer comes from the one node that owns the topic.

8 sample nodes. Nothing leaves your browser.

Plan a domain

Simulation

The planner proposes a tree. You can only approve or reject it; nothing is written before.

Topic Row Level Security for a notes app
  1. Plan
  2. Your approval
  3. Research
  4. Done

Tree preview

Simulation

Every KB is a tree. Open and close branches; click a node to see what it is.

Click a node, or play the tour.

  • Root root
  • Class class
  • Instance instance
  • Part part

Platforms

One account, three places to work

Web

Read and ask your KBs, start research on our servers, approve or reject plans.

Right here, in any modern browser.

Desktop app

Everything on the web, plus:

  • Your own agents. Define agents and run research with them on your computer.
  • Local research. Runs on your own subscription, not on the starter allowance.
  • Remote control from mobile. Start a research on your phone and your desktop runs it.
Get the desktop app

Mobile

Read on the go, define agents and send work to the desktop at home.

Android today; iOS later.

About AgensTown

We build agent tooling on Claude

AgensTown was founded in January 2026 and is a team of five people.

Knowledgebase

This product. Strict, sourced knowledge trees researched by Claude agents. An Ask tool over MCP and an HTTP API lets other agents query a KB. Web, desktop and mobile apps.

Consistent

A researcher agent, in production since October 2026.

Built on Claude

Research runs orchestrator-worker: an orchestrator plans the domain and hands each node to a worker, and every worker is a Claude Code session that researches and writes the node's sourced sections. Other agents reach a KB through MCP.

Claude is made by Anthropic. AgensTown is an independent company, not affiliated with or endorsed by Anthropic.

Contact us at hello@agenstown.com

Start with your Google account

Sign in with Google to read and ask. Cloud research is switched on per account.

Create account Sign in