Welcome back
Your morning brief, your money, and answers grounded in your own accounts.
Your morning brief writes itself here the moment your money connects. Three ways in: accept a family-money share (in your profile — tap the C, top left), upload a bank CSV, or connect your bank — Plaid arrives in days.
Your real brief from your connected Gmail, Tasks and money.
Ask anything — Cecchin knows your rich life and your money, and never moves a cent.
Estimates from RentCast — the address, plus what the building is if you fill that in, sent only when you tap Get value. What you paid, your insurer and your notes stay here. Refreshed at most monthly; an estimate is a direction, not an appraisal. Address suggestions come from Amazon Location Service — inside the same AWS account that runs Cecchin — which sees what you type in the address box as you type it, and nothing else.
Nothing filed yet — inspection reports, insurance policies, move-in photos all live here.
Records for now — photo & PDF upload arrives with the private storage bucket (your go).
Where your reserve, spend, and taxes are heading — projected from your own data once your accounts connect.
Estimate — “free” = your household’s invested covers 25× its yearly spend (the 4% rule), growing at the real return above. Social Security (if you enter it per person) is shown as income after 67; healthcare before 65 — the Medicare bridge — is on you. A direction, not a promise.
Read-only · import a statement · disconnect
Cecchin can see your balances and transactions and can never send or move a cent. The connection stays encrypted in your account.
Export a CSV from your bank and drop it here — it files into your personal money. Re-uploading the same file is free, nothing doubles up. A business statement goes in 📘 My books → 🏦 Business banks, so the two never mix.
Finance and Familia — reading your own transactions in the cloud. They analyse and recommend; they never move money or change a record.
Ask a real "what if…" about your money — Cecchin answers with the tradeoff, a recommendation, and what would change it.
Loading your tasks…
The whole desk, one color per job. Most of these are live now — the two greyer ones arrive with your own data.
In your own words — the life this is all for. Cecchin keeps it in view on your Personal layer.
What the business is really for — the shape of it when it's working. Cecchin keeps this in view on your Business layer.
Bring what Cecchin should know about you — paste notes, or the text from a ChatGPT / Claude / Gemini export. Stored in your account, encrypted and yours to delete.
It's yours. Take it or erase it, anytime.
Loading your inbox…
Loading your calendar…
Loading your contacts…
Two lines about what worked and what didn’t — it goes straight to the person building this.
＋ The fence added in code, after the skill
Arrives as your data connects.
💸 What this costs — every vendor you pay, by calendar month
🔌 API connections — everything this fleet talks to
✎ Typed subscriptions & budget — what no API here can read; entered by you, never a bill
👥 Users — every account, newest first · what it holds and what it has cost in 30 days
⏳ Waiting — the waitlist
Manage — your ops copilot
Metrics
🏦 Bank reads — what Plaid costs, live · balances are billed per call, transactions are not
🧪 Demo — who tried it, and what they say
🤖 AI — the model answering now, then spend by person, by activity and by model
☁️ AWS bill — the invoice itself, month to date
🔌 AI vendors — which companies this deployment may reach. Anthropic answers everything; the rest are a model tournament, one search fallback, and spoken calls.
⚔ Arena — who wins where · avg blind-judged quality 0–10 · ★ champion · beside what that model cost (30d)
🗺 The line of report — who answers to whom; no agent answers to another agent
- Level 0 You Every 0–10, every ideal answer, every ✓ on a proposed move, every blind pick. The only ground truth in the system — no model score overrules it, and once you have graded an agent, machine scores are excluded from its crown entirely.
- Level 1 The orchestrators — Innovation & evolution · Arena They run the agents and are never graded by them: the thing that grades must not be the thing being graded. Innovation & evolution is ONE loop in four moves — Scout finds what the world is doing, Ideas finds where we fall short, Triage decides what is worth doing, Evolve rewrites the words behind the golden gate, and Engineer writes the code. Arena answers a different question: it fights the vendors on real work and crowns which model serves it.
- Level 2 The agents Everything that produces something you read — Doge, Familia, Scenario, the morning brief, mailwatch, autopass, and every agent you build yourself. Each has a skill a rating can rewrite. They report upward through the loop, not to each other.
- Level 3 The deterministic floor Guardians, the shipments scan, and every figure on the Money tab. No model anywhere in these paths, so they cannot invent a number — which is why they sit under everything else rather than beside it.
🧑🤝🧑 Agents — every one in the system and every one a user built, with the LLM that serves it
🧠 How a run thinks — the method is chosen per run, not per agent: the same agent can answer six ways
Versions — every build, grouped by day
Log a decision and what you expect. Later, mark how it went — that's how the judgment gets sharper.
Open
Graded
Anything Cecchin wants to do in your Google account, and anything it decided it shouldn't do alone, waits here until you say so. Nothing runs without your tap.
Waiting
Decided
A spoken call is the one part of Cecchin that doesn't run on Anthropic: your voice and the figures it needs go to the realtime voice provider for the length of the call. Everything typed stays on Anthropic. How this works