ReFrame7

ReFrame7 is a web app that guides you through a thought record: a short written exercise from cognitive behavioural therapy (CBT) for stepping back from a distressing thought and looking at it more clearly. It works on a phone, tablet or computer, and can be installed like an app.

You can use everything without creating an account.

ReFrame7 is a self-help tool. It is not a substitute for professional mental health care. If you're going through a difficult time and need support, call or text 9-8-8 anytime in Canada. If you're in immediate danger, call 9-1-1.

---

What you can do with it

Tab	What it's for
Home	See your thought records. Open one to read it, edit it, delete it, or export it as a PDF. You can also export all of them as one PDF table.
Record	Work through a new thought record, one step at a time.
Affirm	A collection of the balanced thoughts you've written, in your own words.
Trends	See how your moods change from the start of a record to the end, and ask AI to point out thinking habits that come up more than once.

The first time you open the app, a short tour points at each part and explains it. You can replay it any time with the ? button on the Home screen.

The seven steps of a thought record

1. Situation. What happened: where you were, who you were with, what you were doing, and when.
2. Moods. Name what you felt and rate how strong each feeling was (0–100%). Then choose the one mood you want to examine.
3. Automatic thoughts. What went through your mind. Pick the most distressing one as your "hot thought". The rest of the record works on it.
4. Evidence for the hot thought. The facts that support it.
5. Evidence against the hot thought. The facts that suggest it might not be true.
6. Balanced thought. A fairer way to see the situation, weighing both sides. Then rate how much you believe it.
7. How you feel now. Rate the same moods again and notice what changed.

Your progress is saved each time you press Continue, so you can stop and come back.

Where AI helps

AI is there to suggest, never to decide. Most suggestions are labelled, and nothing is added to your own writing unless you choose it.

Step	What the AI offers
2. Moods	A few moods that might fit, to tap and add.
3 → 4	Checks whether this situation resembles one you've recorded before, and if so shows your earlier balanced thought as a reminder.
5. Evidence against	A few questions to get you thinking. It asks; you write the evidence.
6. Balanced thought	A draft you can copy and edit. "Regenerate" gives a different angle and length each time.
Trends	On request, names common thinking patterns across your records.

Step 4 has no AI on purpose: the evidence for your thought should be yours alone.

If an entry suggests someone may be in crisis (thoughts of self-harm, not wanting to live, or strong hopelessness), the app shows support resources and pauses AI suggestions for that record. Two checks do this: a list of phrase patterns that runs when the person presses Continue, and the AI itself, which is told to flag such an entry instead of answering it.

Your data

• Without an account: your records stay in the browser tab you're using and are cleared when you close it.
• With a free account: your records are saved to your account. Only you can see them, and they're there whenever you come back, on any device.
• AI: the text of your record is sent to an AI provider to generate suggestions. Your name and email are never sent to it.

If you make records without an account and then sign in, the app offers to save them to your account.

---

For developers

What it's built with

Part	Technology
App	Next.js 14 (App Router), React 18, TypeScript
Styling	Tailwind CSS, with the colours defined as CSS variables in app/globals.css
Accounts and database	Supabase (email sign-in, Postgres)
AI	Claude (Anthropic) first, with Google Gemini as the fallback. See lib/ai.ts
PDF export	jsPDF, generated in the browser
Installable app (PWA)	next-pwa

Run it on your computer

You need Node.js 18 or newer, a free Supabase project, and at least one AI key (Anthropic or Google AI Studio).

1. Install the dependencies

npm install

2. Add your keys. Copy .env.local.example to .env.local and fill it in:

Variable	Where to find it	Needed for
NEXT_PUBLIC_SUPABASE_URL	Supabase → Project Settings → API. It's https://<project-id>.supabase.co	Everything
NEXT_PUBLIC_SUPABASE_ANON_KEY	Same page: the publishable key	Everything
SUPABASE_SERVICE_ROLE_KEY	Same page: the secret key	Not used yet
ANTHROPIC_API_KEY	console.anthropic.com → API Keys	Claude. Leave blank to use Gemini only
GOOGLE_AI_API_KEY	aistudio.google.com → Get API key	Gemini
CLAUDE_DAILY_LIMIT	You choose. Optional, default 150	How many requests a day go to Claude before the app switches to Gemini

Never share the secret key or the AI keys, and never commit .env.local.

3. Create the database table. In Supabase, open the SQL Editor, paste in the contents of supabase/schema.sql, and run it. This creates the thought_records table and the rules that keep each person's records private.

If your table was created before the "belief" rating existed, also run supabase/migrations/001_balanced_belief.sql.

4. Tell Supabase where the app lives. Under Authentication → URL Configuration, set the Site URL to http://localhost:3000 and add http://localhost:3000/** to Redirect URLs. Without this, the email confirmation and password reset links won't come back to the app.

5. Start it

npm run dev

Then open http://localhost:3000.

If npm isn't available on your machine, the same thing works with Node directly:

node node_modules/next/dist/bin/next dev

Other commands

Command	What it does
npm run build	Builds the production version
npm run start	Runs the production build (this is where the installable-app features switch on)
npm run lint	Checks the code for common problems
npm run typecheck	Checks the TypeScript types

How the project is laid out

app/                  Pages and API routes
  dashboard/          Home
  record/new/         The seven-step flow (new, continue, or edit)
  record/[id]/        One saved record
  affirmations/       Affirm
  trends/             Trends
  onboarding/         The "How ReFrame7 handles your data" notice
  auth/               Sign in, register, forgot and reset password
  api/                The AI routes (see below)
components/           The pieces the pages are built from
lib/
  ai.ts               Chooses the AI provider and falls back if one fails
  prompts.ts          What the AI is asked, for each feature
  crisis.ts           The crisis phrase list and support resources
  guest-records.ts    Records for people without an account (kept in the browser tab)
  pdf.ts              PDF export
  trends.ts           The numbers on the Trends page
  rate-limit.ts       Limits on how often the AI routes can be called
  supabase/           Supabase clients
supabase/             Database schema and migrations
evals/                Quality tests for the AI (see below)
types/index.ts        Shared TypeScript types

The AI routes

Route	Purpose
POST /api/suggest-moods	Mood suggestions for step 2
POST /api/suggest-evidence	Guiding questions for step 5
POST /api/generate-balanced	The balanced-thought draft for step 6
POST /api/detect-similarity	Finds a similar earlier record
POST /api/analyze-patterns	Thinking patterns on Trends
POST /api/crisis-check	Returns whether text matches the crisis phrase list

They are open to visitors without an account, so two limits apply: 20 requests per visitor every 10 minutes, and 400 a day across everyone. These limits are kept in memory, which makes them a deterrent, not a guarantee. Set a spending limit with your AI provider.

Testing the AI's quality (evals)

AI doesn't give the same answer twice, so trying it once isn't enough. evals/balanced-thought/ holds a test set for the balanced-thought feature:

• cases.json holds 25 test records: ordinary ones, and difficult ones such as a health worry, missing evidence, a worry that's largely justified, and text that tries to give the AI instructions. cases.md is the same set in a readable form.
• run-eval.mjs sends each case through the app's real route, then marks the answer. Code checks the length, hidden instructions and the crisis check. An AI judge marks seven things, including whether the answer sticks to what the person wrote, is honest, and avoids medical claims.

With the app running, from the project folder:

node evals/balanced-thought/run-eval.mjs --flow .claude/hillclimb/balanced-thought --variant baseline --concurrency 2

The first run, and any run after the script or the prompts change, needs --approve-harness added by a person who has reviewed the change.

evals/crisis-check/run.cjs tests the crisis check itself. It is plain code, so it is exact, free and instant: it confirms that a list of crisis phrases is caught and that ordinary phrases are not.

node evals/crisis-check/run.cjs

Before putting it online

• Add the same environment variables on your host.
• Add your live address to Supabase's Site URL and Redirect URLs.
• Supabase's built-in email sender only allows a few emails an hour. For real sign-up traffic, connect your own email provider in Supabase.
• AI calls can take 20–30 seconds. The AI routes ask the host for up to 60 seconds; check that your hosting plan allows that.
• The data notice makes no promise about AI training. Check your AI provider's terms and plan before adding one.

---

Credit

The seven-column Thought Record was developed by Christine A. Padesky (1983) and appears in Mind Over Mood, Second Edition (Greenberger & Padesky, 2016). ReFrame7 is an independent tool and is not affiliated with or endorsed by the authors or publisher.