AI that reads your private files.
On your machine. Not the cloud.

Patient charts, client contracts, student records — the documents that can't go to a cloud LLM. Sery reads them on your machine and lets AI find, summarise, draft, and reason across every file, without the bytes ever leaving the storage they live in.

Sery Link · File Preview
orders_2024.parquets3://analytics/orders/2024/
1.4 GB · 4,209,847 rows · 5 cols · Snappy compression·12 KB read · footer only·Resolved in 1.8s
5 columns
ColumnTypeNullsDistribution
order_idBIGINT0.0%4.2M unique · primary key
customer_idBIGINT0.0%2.1M unique · foreign key
amountDOUBLE0.2%$0.99 → $9,847.00 · avg $42.18
statusVARCHAR0.0%3 values · paid · pending · cancelled
created_atTIMESTAMP0.0%2024-01-01 → 2024-12-31 · partitioned
Bytes read this preview · 12 KB0 bytes uploaded · file stays in S3
Local
HTTPS
S3
S3
Drive
SFTP
WebDAV
Dropbox
Az
Azure
OD
OneDrive
Backblaze B2·preset
Wasabi·preset
Cloudflare R2·preset
GCS·preset

Every storage where your private documents live

Local folders, S3 buckets, Google Drive accounts, SFTP servers, WebDAV mounts, B2 buckets — wherever your contracts, charts, workpapers, and records sit today. Add a source in seconds. Read, search, summarise, and reason across all of them from one private catalog.

16 sources: file stores + databases

9 file stores — Local, HTTPS, S3 (+ B2, Wasabi, R2, GCS), Drive, SFTP, WebDAV, Dropbox, Azure, OneDrive — and 7 databases: Postgres, MySQL, Snowflake, ClickHouse, MongoDB, Redis, SQLite. One app reads from every place your private data is scattered today.

Read PDFs, scans, contracts in-process

Native Rust extraction via pdfium for born-digital PDFs; macOS Vision and Windows.Media.Ocr for scanned pages. Excel via calamine. DOCX, PPTX, EPUB via the Pandoc sidecar. The bytes never leave your machine.

Credentials in your OS keychain

AWS keys, Drive OAuth tokens, SFTP passwords — all stored via your OS's native credential store. We never see them. Pre-flight tests them on save so a typo never persists.

Pair multiple machines into one private workspace

One workspace key pairs the laptops, desktops, and servers where your documents live. AI reads and reasons across all of them in one answer. The files never leave the device they live on.

One workspace key

Create a workspace key on sery.ai, paste it into each machine's Connect dialog. The office iMac, the home laptop, the partner's ThinkPad — all reachable from one private workspace.

Documents stay put

Each machine reads its own documents locally. Only the AI's question and the final answer travel — patient files, client contracts, and tax workpapers never cross the network.

One answer, every machine

“Summarise this client's file across my office and home machine” assembles one answer from documents living on both devices — without ever moving them.

How it works

Three steps. The first two run entirely on your laptop. The third is optional and only happens if you explicitly connect to sery.ai.

1

Connect

Click + Add Source → pick where your data already lives (local folders, S3, Drive, SFTP, a Postgres or Snowflake database, etc.) → enter credentials. Sery pre-flights the connection, stores the credential in your OS keychain, and registers the source as a private bookmark.

  • —9 file stores: Local, HTTPS, S3, Drive, SFTP, WebDAV, Dropbox, Azure, OneDrive
  • —7 databases: Postgres, MySQL, Snowflake, ClickHouse, MongoDB, Redis, SQLite
  • —OS keychain credentials, never on our servers
  • —Pre-flight test catches bad creds on the modal
2

Read & search

Sery's in-process Rust extractor pulls text from PDFs (pdfium), scans (macOS Vision / Windows OCR), Excel (calamine), and DOCX/PPTX/EPUB (Pandoc) — all on your machine. Search across every file's content + filenames + columns in one pass. Tabular files still preview without downloading.

  • —PDFs, scans, DOCX, PPTX, XLSX, CSV, Parquet — all read locally
  • —OS-native OCR for scanned pages (no cloud OCR API)
  • —Content + filename + column search across every source
  • —Auto-profile any column (null %, unique, min/max/avg)
3

Ask (optional)

Connect to sery.ai with a workspace key when you want AI that reads your documents — "summarise this morning's patient intakes," "draft a discovery response from these contracts." The agent runs server-side; only the question and the answer travel. Raw files stay on your machine.

  • —Find / summarise / draft / compare / reason in plain English
  • —AI works across every source and every paired machine
  • —Files never leave their storage; only the answer travels
  • —Workspace sharing for up to 5 invited collaborators

Why the network matters

Sery's value compounds with every machine you add. One install is a desktop app. Two are a network. Five are a moat.

Free · runs on your machine

Sery Link, reading every document

  • • Connect every storage where your documents live
  • • Read PDFs (incl. scans via OS-native OCR), DOCX, PPTX, XLSX, CSV, Parquet
  • • Content + filename + column search across every source
  • • Per-file column profiles (null %, unique, min/max/avg)
  • • Format convert (CSV / TSV / XLSX → Parquet) in place
  • • Local MCP — Claude Desktop, Warp, Cursor, Continue
  • • Runs fully offline; no account required
Plus · AI + network

AI across every document, every machine

  • • Find / summarise / draft / compare / reason across your private files
  • • Cross-machine reading and reasoning in one answer
  • • Cross-machine MCP via mcp.sery.ai
  • • Workspace recipes shared across every machine
  • • Schema-change notifications fan out to every endpoint
  • • 5 invited collaborators on one bill (hub-and-spoke)

Sery's cloud is a coordination layer, not a storage layer. It coordinates — identity, catalog, routing, sharing — but never holds your file contents. The atoms (your patient charts, client contracts, student records) stay on the machines that own them. The network is what makes them readable by AI as one.

Plus plan

Use Sery from any AI tool you already love

Sery exposes your workspace catalog as an MCP endpoint at mcp.sery.ai. Drop your API key into Claude Desktop, Warp, Cursor, or Continue and your LLM gets nine read-only tools to browse, search, sample, and SQL-query every file in your workspace — without any of them leaving the machine that owns them.

~/.config/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "sery": {
      "url": "https://mcp.sery.ai/sse",
      "headers": {
        "Authorization": "Bearer sk_live_…"
      }
    }
  }
}
restart Claude Desktop 9 tools loaded
sery · 9 toolsconnected
list_folderenumerate pathscatalog
search_filesfuzzy matchcatalog
get_schemacolumns + typescatalog
sample_rowsfirst N rowscatalog
read_documentPDF/DOCX contentcatalog
query_sqlremote DuckDBquery
query_dataNL queryai
list_datasetsworkspace catalogai
get_dataset_schemadataset detailsai
SQL travels to your machine; only result rows return.
MCP setup runbook (Claude Desktop, Warp, Cursor, Continue)

Want a local-only MCP server that doesn't need a Sery account? Install cargo install sery-mcp and point it at any folder — same nine tools, fully on-device. See open source.

Put AI to work on the files that can't leave.

Free forever with unlimited machines and the full local data agent. $19.99/mo unlocks AI that reads and reasons across your private files — 7-day trial, no card required.

Get started free