WriteFlow for universities
Process documentation instead of detector scores, a server you can run in your own data centre, and AI that students use within clear, enforced limits.
A server you run yourself
The WriteFlow server is a single Node program, shipped with Docker Compose and Caddy. It only ever sees encrypted envelopes, so running it gives the operator no access to students’ texts. The server software will be published as open source.
- Accounts and groups with the roles read, comment, write and manage
- Invitations by e-mail; removing a member gives the group a new key
- Encrypted PDFs per group
- Hardened: limits on size and rate, strict headers, a security self-test
Provenance without a score
WriteFlow records for each character whether it was typed, pasted, quoted or written by an AI, and lets writers explain their pastes as they go. The result is a description that fits the guidelines on AI use of many universities – not a verdict, and without keystroke biometrics.
AI declaration built in
Permission levels per project, document or chapter; a hash-chained audit trail of every AI interaction, exportable as JSON; and a provenance page in Word, PDF and print exports to hand in with the work.
Free choice of provider
Anthropic, OpenAI-compatible services or a local model: universities can point WriteFlow at their own AI proxy.
Coming for universities
- Process evidence with qualified time stamps and an open-source verifier Planned
- Sign-in via eduGAIN / Shibboleth Planned
- A package for universities: data protection impact assessment, data processing agreement, sample texts Planned
- Moodle integration Planned
- Comments and suggested changes by supervisors Planned
Interested in a pilot?
WriteFlow is looking for a writing centre to try the process documentation in practice. Contact details are in the imprint.