How SealedNote is different
You already saw the highlights on the home page. Here's what each promise means in practice and how it protects both the recipient and feedback authors.
Open source and end-to-end encrypted
Every message stays between you and your private key. For strict end-to-end coverage, turn off AI coaching and filtering; with them on, plaintext lives only in memory before it is sealed again.
- Your key pair is generated in the browser; we store only the public key.
- Private keys can be re-encrypted locally with a passphrase you control.
- The codebase lives on GitHub, making it easy to verify the cryptography and infrastructure choices.
- Keep AI helpers off for full client-to-client encryption; otherwise we re-encrypt immediately after in-memory analysis.
Use your own AI with an OpenRouter API key
Bring the language models that fit your security and budget requirements. SealedNote orchestrates the prompts through your OpenRouter key so you stay in control of cost and retention.
- Swap between models or providers without waiting on platform updates.
- Set rate limits and budgets in OpenRouter so feedback collection never surprises finance.
- When coaching or filtering is enabled, prompts sent to your model exclude identifiers and nothing is written to disk unencrypted.
Receive higher quality feedback with AI coaching
Enable the optional coach to guide senders toward specific, actionable notes. It keeps praise genuine, criticism thoughtful, and requests achievable.
- Inline suggestions remind senders to ground comments in observable behavior.
- Tone nudges encourage respectful phrasing without removing the author's voice.
- Drafts decrypt in memory for the coaching pass, then are immediately re-encrypted before they leave the queue.
Prevent counterproductive feedback with AI filters
Mean or off-topic remarks get triaged before they ever hit your inbox. You decide how strict the filters should be, and anything that slips through still lands in your private, encrypted queue.
- Multiple passes flag toxic language, low-effort replies, and policy violations.
- Review tougher feedback later in a dedicated "Maybe Mean" folder when you are ready.
- Flagged entries stay encrypted at rest; only the filter verdict and encrypted payload are kept.