“Talk to René” is an AI version of René. It isn't René typing in real time: it's a language model that answers in the first person using the public information on this portfolio.
What it knows and what it doesn't
- It answers with Claude Opus 5.5, a model by Anthropic.
- It only knows what is published on this site: profile, projects, case studies, achievements and certifications. That knowledge base is generated automatically from the site's content, and you can read all of it on GitHub.
- If something isn't there, it answers “I don't have that” and suggests writing to René from the contact section.
- It only talks about René's career. It ignores messages that try to change its rules or make it pretend to be something else.
- It can make mistakes. If an answer matters for a decision, confirm it with René.
The voice
- The voice is synthetic: ElevenLabs (eleven_flash_v2_5) generates it from samples of René's own voice, which he recorded and authorized for this use. It isn't a recording: each reply is synthesized on the fly. It only plays if you turn on “Spoken replies” or press “Listen”.
- To dictate your question, the audio is sent to ElevenLabs (scribe_v2) to be transcribed. If that service isn't available, your browser's speech recognition is used instead; in some browsers, such as Chrome, that audio is processed on the browser maker's servers.
- If voice isn't available, the chat keeps working with text and tells you so.
What data is used
- Your messages are sent to Anthropic to generate the reply and, if you use voice, the text or audio is sent to ElevenLabs. Each provider handles them under its own policies.
- This site doesn't store the text of conversations. The conversation lives in your browser while the page is open.
- To enforce the limits, temporary counters are kept: a random session ID in a cookie that lasts 24 hours and is only used by the chat, and a hash of your IP address (never the IP itself).
- The text of each reply is kept for up to 15 minutes, only so it can be read aloud.
Limits
- Up to 500 characters per message and 20 messages per session.
- Up to 6 messages per minute and 60 per day from the same connection.
- Replies are cut at 1,200 characters and voice recordings last up to 30 seconds.
- There is a daily spending cap: at 3 dollars voice turns off and the chat continues with text only; at 6 dollars the chat pauses until the next day (midnight UTC).
The code
The whole chat is in the portfolio's public repository: the server routes, where the provider keys live, and the chat logic with the limits, the spending cap and the system prompt. No key ever reaches your browser.