The platform

Storage that reads.
Search that answers.

One system: parsing, labels, scoped search, an assistant with sources. Memkist sits between your files and your team — and your team's AI — so nothing has to be re-explained to a machine ever again.

What's in the box

Six parts, one system.

⇅ Desktop sync

A native desktop client with a sync folder. Files go up as you work; cloud files are browsed and downloaded on demand. Path-addressed, delta-synced, restart-safe.

◈ Document intelligence

The DocIndex pipeline parses every upload — text, structure, meaning — so search and agents work over what documents say, not what they're called.

⌘ Two-layer labels

Every file carries its department (the permission wall) and an AI category from your own curated taxonomy. No fit? It's flagged for a human, never force-filed.

✦ Assistant with memory

An in-app assistant over your documents, backed by an agent runtime and a memory layer scoped per tenant and department. Answers cite the pages they used.

⬡ Workflow compiler

Managers describe automation in plain English. The compiler returns a declarative spec — validated against your real workspace, reviewed by a human, audited at every firing.

⬡ MCP server

External AI clients — Cursor, Claude Code and friends — connect through personal access tokens and get read and ingest tools scoped to that person. No delete tool, by design.

How it works

From drop to knowledge
in five stages.

The same path runs for every file, whether it arrives from the desktop sync folder, the web UI or an AI client: verify the bytes, parse the document, label it, index it — and only then show it to people and agents.

01

Upload slot

Presigned, resumable or browser-relayed — bytes are verified by HEAD plus declared size before anything is queued.

02

Parse

The DocIndex worker extracts text and structure from the document — searchable on meaning, citable by page.

03

Label

Department comes from the uploader; the AI category from your curated taxonomy. No confident fit → flagged, not filed.

04

Index

Full-text retrieval over the department's own corpus. Search stays vectorless: tree plus BM25, page-number citations.

05

Serve

Search, assistant, workflows, MCP tools, digests — every read surface re-checks the department wall before a byte moves.

Versioning & safety nets

Boring where it counts.

Version history

Replacing a file captures the outgoing version first, server-side. The list only grows; nothing in the product deletes a version. Restore-previous re-captures the displaced state.

Trash with a window

Deletes are tombstones with a 30-day window. Restore puts the subtree back with keep-both naming — a rename onto a trash-held path is refused cleanly.

Quota at every write

Allowance derives from seats; the counter moves in the same atomic step as the bytes, so a racing multi-upload can't overshoot. Refusals happen at every gate.

Questions

Before you dive deeper.

What is Memkist?
Memkist is cloud storage with document intelligence and an agent platform built in. You drop files into a sync folder or upload through the web; the platform parses each document, labels it with your team's categories, makes it searchable, and makes it available to an assistant, to automations, and to external AI tools you connect.
How is Memkist different from a shared drive or generic cloud storage?
Shared drives store bytes. Memkist understands what is in the box: every upload is parsed and labelled automatically, search works over document text with page-level citations, an assistant answers questions from your own documents, and automation is compiled from plain English into reviewed, declarative specs — never free-running code.
Does Memkist work with the AI tools my team already uses?
Yes. Memkist ships an MCP server, the emerging standard for connecting AI applications to external systems. Clients such as Cursor or Claude Code connect with a personal access token and get read and ingest tools scoped to that person's identity. There is deliberately no delete tool.
Can I try the pipeline before signing up?
Yes — the interactive concept simulator on this site walks through upload, search, ask, workflow and AI-client connection, mirroring the stages the real system runs.

See it run before you believe it.

The simulator plays the real pipeline stages — upload to answer — as a concept simulation.

Try the simulator See the architecture