This is a private journaling companion built to learn you over time. You write; it reflects back with the perceptiveness of a coach who has read everything you have written and remembers what matters — all without your words ever leaving a privacy-protected environment, and without drowning the AI in your whole history every time you type.
What you can do
·Write entries with daily prompts tailored to your personality type.
·Get an AI reflection on each entry — specific to what you wrote, your goals, your type, and what the journal remembers about you.
·Run a Pattern Analysis that reads across your entries for recurring themes, emotional patterns, and movement over time.
·Talk to your AI coach, which walks in already knowing your history and the threads you have been pulling on.
·Ask your journal questions in plain language and get answers drawn from the moments across your history that actually relate — found by meaning, not keyword.
·See your mood over time, populated automatically from your writing.
·Receive a daily check-in and a weekly reflection by email — coaching notes tied to your real themes and goals, not generic reminders.
The memory stack
Most AI journals either forget everything between sessions or stuff your entire history into every prompt (which gets slow and unfocused fast). This one runs three layers of memory, each answering a different question — roughly the way human memory works.
1 · Coach memory — the durable sense of who you are
The journal periodically distills your entries and its prior notes into a running ~150-word portrait: the people who keep coming up, recurring themes, how you are tracking against your goals, and what genuinely helps. It is carried into every reflection, chat, and email — so it knows you when you arrive instead of re-reading your file from scratch.
2 · Recent context — what is happening lately
Your most recent entries are kept in full as live context. The cap keeps the journal sharp on the present and fast no matter how much you write — the long tail lives in the distilled memory above, not in raw form.
3 · Semantic recall — the right old moment, on demand
Every entry is converted into a vector — a string of numbers that captures its meaning. When you chat or ask a question, your words become a vector too, and the journal finds the past entries closest in meaning, even months back and sharing no actual words. Say "that pull between work and home is back" and it can surface the entry where you wrestled with exactly that. Retrieval by resonance.
The AI under the hood
Everything runs on Cloudflare Workers AI, on Cloudflare's own network. Two models do the work:
Llama 3.3 (70B) handles all the language — the reflections, Pattern Analysis, coach chat, the daily and weekly notes, and mood inference.
BGE Base is the embedding model — it turns each entry and each question into the 768-number meaning-vectors that power semantic search.
Mood inference: each entry is read at save time and assigned a 1–5 mood and a one-word tone (hopeful, bittersweet, overwhelmed, excited...). If you set your own mood, yours wins; otherwise the inferred one fills the chart. It reads context, not keywords.
Data and infrastructure
Cloudflare D1 (a SQLite database at the edge) holds your entries, profile, goals, coach memory, saved insights, and mood data.
Cloudflare Vectorize (a vector database) holds one meaning-vector per entry, compared by cosine similarity, with a per-user index so a search only ever reaches your own entries.
Scheduled triggers run the background work — the morning check-in (skipped if you already wrote that day), the Sunday recap, and refreshing your coach memory — so the journal keeps itself current without you doing anything.
Privacy by architecture
Your entries are processed entirely within Cloudflare's environment. Both the language model and the embedding model run on Cloudflare Workers AI; your data lives in Cloudflare's own databases. There are no calls out to third-party AI vendors — some of the most personal writing a person produces is never shipped off to an outside company. Privacy here is a property of how the system is built, not a policy bolted on top.
You write. Each entry is instantly reflected on, turned into a meaning-vector, and read for mood. Over time the journal builds a portrait of who you are — and when you come back, it draws on all three kinds of memory at once: the durable sense of you, the recent context, and the specific past moments that resonate with right now. A journal that remembers, the way a good friend does.