Sylou
Process-based AI-use transparency. An open-source writing environment that makes human-AI collaboration visible and shareable: across research, ideas, structure and prose.
[!TIP]
New here? Learn more at meetsylou.eu, share feedback to help us out, or sign up for occasional updates.
Sylou is a text editor that helps you write with AI assistance and analyses your collaboration with AI using a standardized framework that looks into the processes of writing: where ideas came from, what got structured by whom, how the prose itself was shaped, etc.
As it's becoming ever more difficult to understand how AI was used to produce a writing, with standard detection tools being occasionally unreliable and missing the full picture, Sylou aims to flip the problem: helping users gather evidence and proactively demonstrate how they have used AI, using a standardized framework that looks into the processes of writing, producing visual and shareable reports along the way.
Sylou's whole point is to make human-AI collaboration visible for users willing to show this transparency. It's designed for writers, students, educators, researchers, and anyone who cares about AI-use transparency and is looking to proactively demonstrate how they have used AI, in a clear, visual and insightful way.
Additionally, it can further be used in an educational context as a means to promote one's own AI-use awareness, and help serve as a framework for AI-use.
https://github.com/user-attachments/assets/da67993d-da0b-440f-bb94-2fb8d1c3175a
The FLOW methodology
Sylou analyses human-AI collaboration across four dimensions thanks to a structured analysis framework, "FLOW", that maps how AI assistance and human work fit together across the full lifecycle of the writing process.
Each dimension is assessed independently, producing a dedicated and sharable FLOW report, as well as a FLOW collaboration style for your work: one of sixteen possible types, not unlike personality-type frameworks at the level of a piece of work.
FLOW describes the writing process and, in doing so, has no "pass" or "fail", but instead a structured analysis of who contributed to what and how. And it exists alongside your own self-evaluation, if provided.
The four FLOW pillars are:
F — Find: research and critical thinking (research it)
L — Lead: ideas and opinion (form ideas about it)
O — Organize: structure and argumentation (structure it)
W — Write: writing and authorship (write it)
The transparency architecture, covering all the means, whats, whys, hows to achieve this goal, is detailed in a separate and dedicated document, in case you're interested or plan to contribute: TRANSPARENCY.md.
What's in the box
- Writing editor with AI assistance that is context-aware of the work (Ask, Propose, Expand, Simplify, Rephrase, Translate, Fix grammar) and with per-paragraph provenance tracking.
- AI chat assistant that has awareness of the work being written, and with multiple threads per project, document upload capabilities, integrated web search, and prompt caching for token cost efficiency.
- Creative writing tools aimed for fiction writers: world-building tool, character diagram and timeline.
- Other convenience functionalities such as virtual post-its and chat messages bookmarks.
- FLOW analysis that produces a structured report analysing the four pillars of the FLOW methodology.
- FLOW collaboration style based on the FLOW analysis, describing how you worked with the AI, akin to personality-type frameworks at the level of a piece of work.
- Self-assessment options for users to record their own feedback (both before and after the FLOW analysis is provided) on the collaboration alongside the FLOW analysis.
- Work replay functionality and quantitative data for further evidence of the involvement of the user to the work.
- Shareable FLOW reports comprising the above elements, with a public URL, a QR-coded badge, and machine-readable AI-origin markers.
- Tutor mode that restricts the AI to guidance rather than direct generation (for educational contexts).
- Selectable LLM for both the chat (where effort is also selectable) and the FLOW analysis separately, between Mistral (incl. GLM), Claude and DeepSeek.
Sylou is built on React + TypeScript on the frontend, Supabase (PostgreSQL + Edge Functions) on the backend, and uses LLMs via API.
Try it
There's no public hosted instance. The sylou.eu instance is maintainer-only, sign-ups are disabled. To use Sylou, deploy your own instance.
It is open-source software designed for self-managed deployment on Supabase plus a static hosting provider for the frontend (Vercel, Cloudflare, or similar), using your own accounts. You bring the cloud accounts, the LLM API keys, and the domain; Sylou provides the application code.
Setup is documented in docs/setup.md. At a high level, you'll need:
- A Supabase project (free tier works for personal use)
- API keys for at least one LLM provider (with either Mistral or Anthropic models being required for the FLOW analysis). For the FLOW analysis, GLM 5.3 (through Mistral API) or Claude Sonnet 5.5 from Anthropic API are recommended options
- A few environment variables
- About 30-60 minutes for first-time setup
The same self-hosted instance can be used by a single person, a small team, an academic department, or a community; whatever fits.
Self-hosters can set per-user token budgets via SQL; see docs/setup.md for details.
How it works (briefly)
When a project is analysed, the FLOW prompt examines the editor content, the creative writing tools, the chat history, the AI interactions, the provenance tags metadata and the imported documents to evaluate each of the four pillars and assign a collaboration style.
The result is rendered as a FLOW report, which can be kept private or shared via a public URL.
A more thorough explanation of the transparency mechanisms is in TRANSPARENCY.md.
The recommended LLM and settings to use for the FLOW analysis can be found in docs/model-configuration.md.
Open source
Sylou is open-source software released under AGPL-3.0. The FLOW methodology is published separately under CC BY-SA 4.0. Anyone can inspect the code, modify it, self-host it, or adapt the methodology under the terms of those licenses.
A precise mapping of which files fall under which license is in LICENSING.md.
The Sylou name and logo, and the FLOW Pledge Badge format, are governed by a separate Trademark Policy. In short: the Sylou name is not authorized on substantively modified or competing versions. Most contribution scenarios are entirely unaffected by the trademark Policy: PRs, branches, forks, distribution packages, self-hosted deployments, and educational use are all fine, and the bias of the trademark policy is toward permission. The policy only comes into play for cases that could mislead users about what they're using.
Contributing
Contributions are welcome: code, documentation, translations, FLOW methodology observations, bug reports, ideas, conversation. Please see CONTRIBUTING.md if you want to get involved.
Sylou uses the Developer Certificate of Origin for code contributions (git commit -s), and a handful of issue templates make it easy to file structured bug reports, feature requests, or FLOW methodology observations.
A note on pace: Sylou is currently maintained by one person on limited time. Reviews and responses may take longer than on a commercial or larger project. Your patience is appreciated.
Transparency and the AI Act
The project's documented analysis on the EU AI Act's transparency requirements (most notably Articles 50(1) and 50(2)) is in docs/ai-act-position.md. It's the project's documented position, not legal advice, and self-hosters operating Sylou for other users have their own obligations to consider.
Security issues should be reported privately: see SECURITY.md.
Status
Sylou is in active development. Features may change, bugs exist, and you should keep your own backups of important work.
It's been extensively tested with small to medium-sized works, and is being tested with large-scale works (e.g. novels).
It's provided as-is, without warranty of any kind. That said, it's used daily by the maintainer and is intended to be useful.
Roadmap
Do take a look at our Tally questionnaire, linked at the top of this file, which includes a dedicated question for feedback on possible future functionalities. The maintainer's personal priorities include the addition of new set-ups for more flexibility (including the possibility to run Sylou locally or with self-hosted open LLMs), as well as means to further enrich the FLOW reports.
Besides what's listed there, we're of course open to suggestions and feedback. See also CONTRIBUTING.md.
Note on how it was built
Sylou's transparency mechanisms and FLOW methodology were designed by the maintainer, with the code largely written with AI tools under the maintainer's direction and review: initially Cursor (late 2025-early 2026) for a first, fully offline experimental prototype (separate repository); then Lovable (March-June 2026) for a scaffold of this codebase; since then Claude Code, every change reviewed by the maintainer and backed by tests, CI and DCO sign-offs.
Learn more
- Project landing page: meetsylou.eu
- About Sylou and FLOW: linked from the landing page, plus the documentation in this repo
- Feedback or questions: [email protected], or open an issue
