Nothing leaves the workload
The model, the document parser and the search index all run inside your rented machine. No third-party AI APIs — your files are never sent to anyone.
Co-Desk rents you a dedicated GPU, syncs your cloud files into it, and runs an open AI model right next to them. Ask questions, search hundreds of files, generate Word and Excel documents — and nothing ever leaves your workload.
The model, the document parser and the search index all run inside your rented machine. No third-party AI APIs — your files are never sent to anyone.
Pick from open-weight models sized to your GPU — from Gemma 3 to gpt-oss 120B and Qwen3 235B. The weights run on your hardware, under your control.
Attach a volume and your documents, search index and models survive every redeploy — switch GPUs in minutes without losing your workspace.
Link Dropbox in one click, or add any S3-compatible bucket or SFTP server. Your files sync into the workload automatically, every 10 minutes.
Choose your hardware tier and the model that fits it — the form only offers models your GPU can actually run. Watch the live progress panel while everything installs.
Open Co-Desk in your browser, pick your synced folder, and ask. Answers are grounded in your files with real retrieval — not guesses.
Not a chatbot with a file upload — a workspace that indexes, searches, reads and writes your documents.
Semantic vectors plus keyword search, fused — finds the right passage even when you don't remember the wording.
PDF with layout and tables, Word, Excel, PowerPoint, code and text — parsed with document-understanding models, not naive text dumps.
Ask for a report or a spreadsheet and download the .docx or .xlsx it writes — with headings, tables, formulas and charts.
Dropbox, S3-compatible storage and SFTP today — your folders appear inside the workspace, refreshed every 10 minutes, credentials encrypted at rest.
Deploy Co-Desk on a dedicated GPU in minutes. Per-second billing, no subscription — terminate whenever you're done.