◆

VaultPDF

The Connector Room

Use the AI you already pay for. On your files. Here.

When do you need a key? Only when you want a chat model you already pay for to drive paper jobs (score, redact, extract, merge). Clip Lab does not need a key — open the video door for shorts packs and manuals without pasting anything.

For this room: paste your provider key, drop files, say the finished result. The key goes to your provider only. This page does not store it on AKD servers. Hermes and other agent OS tools: use AGENT_OPS.txt for operation language (results only, no engineering).

Step 1 — Connect your AI

Connect your AI

Works with any provider that allows browser-direct API calls. Pick a preset below to fill the endpoint automatically, or type your own. Your key goes to your provider only — this page never sees it.

Provider preset — fills endpoint automatically

API key

API endpoint (messages URL)

Remember on this device (stores in localStorage — uncheck on shared computers)

Your key goes to your provider only — directly from your browser, never through ours. This page never receives it, never logs it, never proxies it. The moment you close this tab, if "Remember" is off, it's gone.

Step 2 — Load your files

Your working files

Drop the files your AI should work on. They stay in this browser session — the AI receives only small distillates (counts, excerpts, summaries) to work with, never the raw file contents beyond bounded snippets.

no files loaded yet

Drop files here or click to choose — PDF, CSV, TXT, DOCX

Step 3 — Tell your AI what to do

Plain-language instruction

Tell your AI what you need in plain language. It will call the tools it needs — you watch it work and see the results as they land.

Waiting for your first message…

Results land here as downloads when ready. File bytes are not bulk-uploaded to a second document cloud; the chat model only gets the small excerpts needed for the job — distillates (line counts, excerpts up to 500 chars, match lists). The AI calls tools by name; the tools run locally.

What your AI can do here

Available tools

score_resumeScore a PDF resume against a job posting, return keyword gaps and format flags

pdf_mergeMerge multiple PDF files into one, in the order you name

pdf_splitExtract specific pages from a PDF (e.g. pages 1-3 and 7)

pdf_rotateRotate all pages in a PDF by 90, 180, or 270 degrees

pdf_numberStamp page numbers onto a PDF

doc_findSearch a text or CSV file for lines matching a word or phrase

doc_extractPull every matching line from a large file, export as a download

doc_profileProfile a file: line count, size, density, longest line

doc_uniqueCount distinct values in a CSV column, ranked by frequency

doc_columnSum, average, min, max for any numeric column in a CSV

doc_entitiesPull every email, phone, dollar amount, and date from a document

docx_fillFill a Word template's {{TOKENS}} with values you name

What's not here yet

OpenAI direct keys: OpenAI's own API blocks browser-direct CORS. Use OpenRouter instead — it accepts OpenAI models with a browser-direct endpoint, and the preset above wires it correctly.
Gemini: Google's browser CORS policy varies by API version. Not confirmed working yet — named here, not hidden.
Video tools: The Binder room's video processing is canvas-based and cannot be driven by text commands yet. Drop a video in the Binder directly.
Raw file reading by the AI: The AI receives distillates, not raw file bytes. This is the design: the AI cannot leak what it was never given. If you want the AI to read a specific excerpt, say "show me the first 20 lines" and the tool will pass that bounded excerpt.

The deep PDF tools (full typeset report, batch mode) need a pass. See passes →