Run AT-1 where your data already lives
A desktop app, an MCP server for your AI agent, and a Python SDK. The encoder stays on your machine — your files never leave it. We mark what's shipped and what's still in progress, honestly.
Try a Living Database™ demo
Before you install anything, open a real AT-1 Living Database: a whole dataset plus its query engine in one .html file. Searchable offline, nothing uploaded.
AT-1 Desktop
Standard mode for everyone, and an Advanced mode exposing every codec, query, bundle, decompress and integrity feature — plus turning a CSV, a PDF/Word doc, or a whole bundle of documents into a searchable file. Your files never leave your machine.
One self-contained installer per OS — the compression engine is bundled, so there's no Python or setup. macOS and Linux builds come off the same source; where a platform isn't published yet we say “available at launch” rather than hand you a dead link.
Windows
AvailableSelf-contained installer (.exe). Standard + Advanced modes, the engine is bundled — no Python, nothing to configure.
macOS
AvailableUniversal build (.dmg) — runs natively on Apple Silicon and Intel. Unsigned for now: right-click → Open the first time.
Android
Coming soonA mobile companion is on the roadmap — not yet shipped.
Every installer is verified-lossless and runs entirely offline — the engine is bundled, nothing is uploaded, and there's nothing to configure.
Connect your account
On first run, open the Connect account panel and paste an API key from your dashboard (under API Keys). The desktop app meters usage against that account just like the CLI — creating archives requires a free connected account, and the free tier covers a one-time 100 GB-compressed trial (free account, no card) as a credit, not a hard cap. Once it's used: add a card and usage continues at the metered TB-under-management rate; with no card, encoding is blocked (card_required) until you add one. Decoding existing .at1 files is always free. See free tier & limits.
AT-1 MCP server
Give Claude Desktop, Cursor, ChatGPT and your IDE four tools to compress, inspect, query, and tamper-check AT-1 archives — a thin, robust wrapper over the tested at1 CLI, running locally over your own files.
Install & run
One native binary, no Python — the MCP server is built into the CLI:
npm install -g @tinyfiles/cli # native binary, no Python at1 mcp # stdio server # or with no global install: npx @tinyfiles/cli mcp
Decoding/verification never needs an account; encode & query are metered against the API key the host process supplies.
Claude Desktop — claude_desktop_config.json
Packaged command, no paths:
{
"mcpServers": {
"at1": { "command": "at1", "args": ["mcp"] }
}
}Or with no global install (npx):
{
"mcpServers": {
"at1": {
"command": "npx",
"args": ["@tinyfiles/cli", "mcp"]
}
}
}The four tools
| Tool | What it does |
|---|---|
at1_compress | Verified-lossless compress; queryable=True for the in-place-queryable codec. |
at1_info | Codec, schema (rows / columns / types / row-groups), integrity status. |
at1_query | SELECT … WHERE over a queryable .at1 — reads only the blocks the query touches. |
at1_verify_integrity | Recompute the embedded SHA-256 and confirm the bytes are unchanged. |
Then ask your agent: “Compress this CSV with AT-1 as a queryable archive, then find the rows where c0 is between 100 and 200,” or “Verify this .at1 hasn't been tampered with.” The server runs locally — paths are your own files, nothing is uploaded.
CLI & browser decoder
The full product — every codec, the at1 CLI, and the queryable read-side API — installs from npm as a self-contained binary (no Python required). The decoder is open and runs anywhere, including in your browser.
CLI
AvailableOne install — a self-contained binary with all codecs and the CLI, no Python required. Encode, decode, and query .at1 files in place.
npm i -g @tinyfiles/cli # a self-contained binary — no Python requiredRead the SDK docs
Try it in your browser
AvailableThe decoder is compiled to WebAssembly, so you can decode and verify an AT-1 archive without installing anything — it all runs client-side.
Open the browser demo