FAQ · journal bot

How do I sign in to the website?

Two ways. Both end up on the same place — your timeline.

Telegram button — go to /login and click "Sign in with Telegram." Works in any browser where Telegram is also signed in.
6-digit code — for any device. Send the bot /login; it replies with a code; paste it on /login/code. Code is valid for 5 minutes and works exactly once.

If you're brand new and don't have an account yet, head to /start.

How much does it cost?

Free during beta. We'll publish pricing on /pricing before anyone gets charged.

Where does my data live?

On the same server you sign in to. Your audio files, your photos, your transcripts, your embeddings — all on one machine, not spread across cloud providers. You can take it all with you using the export button on the settings page.

Can the developer read my journal?

Depends which encryption mode you chose:

Standard mode: the operator could, if they wanted. The master key that wraps yours lives on the server. It's encryption against DB leaks, not against a curious operator. The explicit commitment is: not to read.
Passphrase mode: at rest, no — the decryption key is derived from your passphrase and is never stored. When you sign in to the website, your key sits in RAM while you're active (so workers can transcribe your audio), and is wiped when you sign out. The only window where someone with root access could read your data is while you're actively using the site.

Same in both modes: no third parties — no AI vendor, no tracker, no analytics service. All models run locally.

Is the AI sending my voice to OpenAI / Google / anyone else?

Three local AI models run on the server: Whisper turns your voice into text, Qwen analyses sentiment and answers your questions, nomic-embed makes the entries searchable. The current configuration does not call out to any third-party AI API (OpenAI, Google, Anthropic). You can verify which models are running at /status.

How do I delete my account?

Two ways. Both fully delete everything — entries, audio, photos, embeddings, reminders, your encryption key, your settings, your account row.

From Telegram — send the bot /forgetme CONFIRM. No website needed.
From the website — Settings page → "Delete account" → type DELETE → click "Delete my account". Your session cookie is cleared automatically.

After deletion, sending /start to the bot creates a brand-new account from scratch. The audit log keeps a record of the deletion event but with the user_id anonymised.

Standard vs passphrase encryption — what's the difference?

On first sign-in to the website you're asked to pick. Both encrypt your transcripts, summaries, embeddings, tags, reminders, mood, and sentiment score — the difference is who holds the key.

Standard (recommended). Per-user encryption key generated by the server, wrapped with a master key the operator holds. A database leak alone yields ciphertext. Operator could decrypt if they wanted to (full disclosure on the privacy page). Nothing for you to remember.
Passphrase. Key derived from a passphrase only you know — never sent to the server in storable form. The operator cannot decrypt your data at rest. If you forget the passphrase, your data is gone forever. No reset, no recovery.

You can switch between the two modes any time from Settings → Encryption. Switching re-encrypts everything — can take a few minutes if you have many entries.

What if I forget my passphrase?

You lose access to everything sealed with that passphrase: transcripts, summaries, tags, embeddings, mood. Original audio is also unrecoverable (sealed with the same key). Metadata (dates, duration, language) stays visible — but the content is gone. No recovery. That's why standard mode is recommended for most people — passphrase mode is for those who accept that risk in exchange for the operator not being able to decrypt.

Can I import old recordings?

Yes. Head to /import and upload one or more files (audio or photo). Each one gets transcribed and enriched as if you'd sent it to the bot. Works with files up to ~500 MB; large .wav files transcode to Opus to save space.

What are "on the horizon" goals on /stats?

Things you mentioned in your entries that the AI flagged as aspirations, trips, dates, or goals. You can also add your own directly. Each one shows how many entries mention it (semantic search, not exact-word match). You can mark them done (✓) or dismiss them (✕). Done goals move to /goals/done.

How do I know the system is working?

Check /status — a public page showing whether the database, AI models, and each background worker are healthy. The raw JSON is at /health. No auth required for either; transparency is the point.

Can I run my own copy?

Not today — the codebase stays private while the model is being evaluated (see the encryption question above). It may open up to the public later; if self-hosting matters to you, get in touch and we can talk.

What if the AI gets something wrong?

It will sometimes — especially with proper nouns, foreign words, or noisy recordings. The original audio is always kept on the server (not just the transcript), so you can listen back to what you actually said. Sentiment scores and tags are best-effort hints, not ground truth.