Give every file one of two homes
Most files an AI session leaves behind are working files: drafts, exports, one-off scripts. They belong in the project folder, beside the code they support, and they can go when the work ships. A smaller group is knowledge: the decision you made, the approach that worked, the link you will want again. That group needs a second home, one built for finding things later.
Read each file before you file it
Checking a file is the real work, and ordinary apps fit poorly here: a text editor shows punctuation, a browser shows only the result, an IDE shows an entire development environment. For a calmer read, Artifact Desk opens and edits the files AI tools make. Markdown becomes a calm document with the source one click away, nested JSON unfolds into a labeled tree, and CSV exports become a sortable table. The viewer is free, needs no account, has no file cap, and includes a 14-day Pro trial. Pro covers edits, saving, conversion, and printing.
Keep the reading local
The same viewer is a local desktop front door for files created by Claude Code, Codex, Cursor, Antigravity, and technical collaborators. Files open through Open Anything, drag and drop, a pasted path, a double-clicked associated file, or a local coding agent calling the operating system. It runs locally on macOS and Windows, keeps visual views as projections of the real file rather than detached cloud copies, and never uploads artifact contents. It is not an IDE: no dependencies, no development server, no replacement for a coding environment. Unknown files fail safely, with unknown text opening as text and unknown binary limited to metadata and bounded hexadecimal inspection (the AI artifact viewer page).
Decide what is worth keeping
Ask one question of every file: would I want this in three months? A decision and the reason behind it passes. So does a short playbook, a working configuration, or a link you would otherwise lose. Regenerable drafts do not; they stay with the project until it ships. KnowMan keeps this kind of material as memories: claims, decisions and playbooks.
Save links and notes where search works by meaning
A knowledge base only helps if finding beats re-reading. KnowMan saves articles and links, reads and tags them, and answers questions with citations. Search matches meaning, not just keywords, so a question in your own words can find a page saved months ago. The Chrome extension saves pages as you browse. For the wider picture, see a personal knowledge base guide and the AI knowledge base guide.
Let your agents read what you kept
Keeping knowledge pays off when your tools can reach it. KnowMan ships an MCP server in two forms: a local stdio server started with npx -y @knowman/mcp on Node 22 or newer, and a hosted remote server at https://api.knowman.io/mcp. Agents gain tools such as kb_search, kb_ask, kb_get_item and kb_save, so a coding session can search a saved runbook or save a URL mid-task. Documented configurations exist for Claude Desktop, Claude Code and Cursor. Start with the MCP server knowledge base guide or knowledge base for AI agents.
Where KnowMan fits
A file viewer answers what the AI just wrote. A knowledge base answers what you know. KnowMan is built for the second question: it saves articles and links, reads and tags them, keeps memories as claims, decisions and playbooks, and answers questions with citations. Agents reach the same material through the MCP server. The Free plan includes five Ask questions a day, Pro is $12.99 a month or $99 a year, and a 14-day Pro trial needs no credit card.
Artifact Desk covers reading those files in read and keep the files your AI writes.
Questions people ask
Should I keep every file my AI writes?
No. Keep working files in the project folder until the work ships, and move only durable knowledge, such as decisions, playbooks and reference links, into a knowledge base. Files you can regenerate on demand rarely need a second home.
Can a coding agent search what I saved?
Yes. KnowMan's MCP server gives agents tools such as kb_search, kb_ask and kb_save, locally over stdio or through the hosted remote server. See the MCP server guide.
Does KnowMan connect to Claude Code or Cursor?
Documented configurations exist for Claude Desktop, Claude Code and Cursor, describing how to point them at the KnowMan MCP server. The Claude Code knowledge base guide covers the setup in detail.
What does KnowMan cost?
The Free plan includes five Ask questions a day. Pro is $12.99 a month or $99 a year, and every account can start with a 14-day Pro trial, no credit card required.