Claude Cowork
Work with Claude on sensitive documents without him reading them.
The Safe-Doc skill for Claude (Cowork, Desktop, Code) pseudonymizes your files before they enter the conversation. Claude works on tokens such as [PERSON_1], and real values are restored on your machine. Processed on our servers in Europe, purged after processing, with no third-party AI.
How it works
- Download the skill from your Safe-Doc workspace: your API key is already included.
- Install it in Claude and allow
safe-doc.ai: two minutes, once. - Hand Claude a file: he does not open it. A skill script sends it to Safe-Doc, and Claude only receives the pseudonymized copy.
- Get the clear result back: real values are restored locally, on your machine.
Why it's safe
Claude never reads the original file. A skill script sends it to Safe-Doc; only the pseudonymized version enters the conversation.
The mapping table stays with you. The link between each token and its real value is kept on your machine, and restoration is offline: even the final result does not go back through a server.
Stable pseudonyms. [PERSON_1] means the same person across all files in a session: Claude can follow who is who without knowing who they are.
« Zero third-party AI » applies to pseudonymization: no AI provider sees your content during that processing, done on our servers in Europe. After that, Claude only sees the pseudonymized version.
User guide
Four setup steps, then day-to-day use.
Before you start
- Claude (Cowork, Claude Desktop or Claude Code) with code execution enabled.
- A Safe-Doc account: 10-day free trial, no credit card, or a paid plan.
1. Download the skill
Sign in to Safe-Doc, open the Integrations tab and click « Download the Safe-Doc skill ». The safedoc-anonymizer.skill file already contains your API key: do not share it.
2. Install it in Claude
In Claude, open Customize → Skills and upload the .skill file. The skill is available from the next conversation.
3. Allow Safe-Doc (required)
The skill runs in Claude's sandbox, where Internet access is limited. Add safe-doc.ai under Settings → Capabilities → Code execution and file creation → Domain allowlist. Without this step, Claude returns a 403 error.
4. Use it
Point Claude to a file (PDF, Word, spreadsheet, image or text) and ask for example: « Anonymize this contract with Safe-Doc then summarize it. » Claude pseudonymizes the file, works on the copy, then restores the result locally: the final document with real values is saved in your working folder.
Do not paste sensitive document content into the conversation: point to the file and let the skill handle the rest.
Troubleshooting
- 403 or « domain not allowed »:
safe-doc.aiis not on Claude's allowlist (step 3). - Skill does not trigger: say « with Safe-Doc » in your request, and check it is enabled under Customize → Skills.
- Authentication error: your API key may have been regenerated. Download the skill again from the Integrations tab; it includes the updated key.
- Need help: contact us or write to contact@safe-doc.ai.
The skill downloads directly from your Safe-Doc workspace, API key included. No account yet? The trial is free for 10 days.