Thoughts
A private, single-user personal operating system for journaling, daily execution, structured reflection, and AI-assisted self-review.
Thoughts brings together a structured journal, task and routine management, an insight library, behavioural exercises, AI-assisted capture, semantic search, retrieval-augmented generation (RAG), and an MCP server that lets an external assistant work with the user's private data.
What it does
Thoughts is organized around four connected workflows:
- Capture - Write structured thought cards, use AI to turn raw text into a completed journal entry, and record ideas from books, articles, podcasts, or principles.
- Execute - Plan a day, manage recurring routines, track check-ins, and use the agent control center to update tasks.
- Reflect - Review moods, completion trends, behavioural activation activities, worry experiments, and AI conversation summaries.
- Connect - Search the entire personal archive semantically, ask grounded questions about past entries, and connect lived experiences to source ideas.
Features
Journal and reflection
- Create, edit, and delete thought cards with titles, categories, mood scores, summaries, long-form notes, tags, and concept tags.
- Link thoughts to ideas in the insight library and record evidence of applying those ideas.
- Review entries on a monthly calendar and in an archive of mood- and category-aware cards.
- Add day intentions, end-of-day notes, mood ratings, and timestamped mood/energy/focus check-ins.
AI smart capture
The Smart Capture workflow accepts raw journal text and uses Gemini to infer:
- Title, category, mood, summary, and long-form body
- Tags and concept tags
- Related insight-library ideas
- The reflection explaining how an idea appeared in real life
The generated fields are placed into the journal form for review before saving. API errors, rate limits, overloaded models, token limits, and parse failures are surfaced with actionable feedback.
Tasks and routines
- Create date-specific tasks with priorities, tags, notes, and statuses:
todo, in_progress, done, or skipped.
- Create recurring task templates with weekday schedules, start/end dates, priorities, tags, notes, and active/inactive controls.
- Apply active recurring templates to a date while preventing duplicate instances.
- Roll unfinished tasks forward to another date.
- Track daily completion percentages, 30-day trends, monthly breakdowns, current streaks, and best days.
AI task agent
The AI Task Control Center provides a focused interface for reviewing and changing tasks for a selected date. It can load daily and recurring tasks, create tasks, update task status or details, delete tasks, and apply recurring routines through the application's authenticated APIs.
Insight library
- Store sources such as books, articles, podcasts, and standalone principles.
- Capture individual ideas from each source.
- Track idea depth through
understood, noticed, applied, and internalized.
- Link journal thoughts to source ideas and update progress as understanding becomes practice.
Behavioural and cognitive exercises
- Behavioural Activation: record activities with before/after ratings for depression, pleasure, and achievement, then save them to the journal.
- Worry Postponement: complete a structured seven-day module with worry definitions, evidence for and against, scheduled thinking time, postponed items, experiment-day observations, and outcome comparisons.
RAG and semantic personal search
Thoughts materializes supported journal data into RAG documents and stores chunked embeddings in PostgreSQL with pgvector.
- Index thoughts, book ideas, conversation summaries, behavioural activation entries, day notes, and daily rollups.
- Re-sync documents and optionally force re-embedding when source content changes.
- Configure enabled document types, chunk size, overlap, retrieval counts, custom grounding instructions, and the selected language model in Settings.
- Search the archive using vector similarity with a text-search fallback.
- Augment searches with temporal expressions such as "today", "this week", and "last month".
- Ask questions through grounded RAG generation; responses include provenance excerpts and return "I don't know" when the indexed context does not contain the answer.
- Show live RAG context while writing when live context is enabled.
MCP integration
Thoughts includes a JSON-RPC MCP server at /api/mcp, secured with an API key and configured for one MCP user. Compatible assistants can use tools to read and update the personal workspace, including:
get_daily_summary, get_thoughts, and get_conversation_logscreate_thought and create_conversation_logget_tasks, create_task, update_task_status, update_task, delete_task, and roll_forward_tasksget_recurring_tasks, create_recurring_task, and apply_recurring_tasksget_behavioural_activation_entriesdelete_conversation_logsearch_journal for semantic search across the indexed archive
The server implements MCP protocol version 2024-11-05, exposes initialize, tools/list, and tools/call, and keeps MCP data scoped to the configured user.
Conversation archive
External assistants can save structured conversation summaries containing a date, title, topics, insights, action items, and optional mood context. These summaries remain separate from normal thought cards while still being available to RAG search.
Reports, backup, and portability
- Export daily reports as JSON or CSV.
- Export monthly reports as JSON with grouped day-by-day summaries.
- Export a full JSON backup of thoughts, tasks, routines, and related personal data.
- Restore data from a previous JSON backup.
PWA and operational tooling
- Installable PWA manifest with app icons.
- Service worker and offline fallback page.
- Health endpoint for checking database and Gemini configuration.
- Global semantic search and API-backed retrieval components.
- User-configurable AI, RAG, live-context, and task-agent settings.
Routes
| Area | Route |
|---|
| Public landing page | / |
| Authentication | /login, /register, /forgot-password, /reset-password |
| Journal dashboard | /dashboard |
| Daily execution | /dashboard/today |
| Recurring tasks | /dashboard/tasks |
| Completion analytics | /dashboard/completion |
| Insight library | /dashboard/insights |
| AI task control center | /dashboard/agent |
| Conversation archive | /dashboard/conversations |
| Behavioural Activation | /dashboard/activation |
| Worry Postponement | /dashboard/worry-postponement |
| Settings | /dashboard/settings |
API and integration endpoints
POST /api/smart-capture - Convert raw text into structured thought fields with Gemini.GET /api/retrieval - Retrieve relevant indexed excerpts.POST /api/rag/generate - Answer a question using grounded retrieved context.GET|POST /api/rag/documents - Inspect, synchronize, and embed materialized RAG documents.POST /api/mcp - MCP JSON-RPC endpoint for external assistants.POST /api/conversations/create - Authenticated integration endpoint for conversation summaries.POST /api/auth/forgot-password - Request a password reset link (always returns a generic response).POST /api/auth/reset-password - Validate a reset token and set a new password.GET|POST /api/backup - Export or restore a full JSON backup.GET /api/reports/daily - Generate a daily report.GET /api/reports/monthly - Generate a monthly report.GET /api/health - Check service configuration and database health.
Tech stack
- Framework: Next.js App Router
- Language: TypeScript
- UI: React
- Styling: Tailwind CSS
- Database: PostgreSQL via
pg
- Vector search: PostgreSQL
pgvector
- AI: Google Gemini for embeddings, smart capture, and grounded generation
- Authentication: Signed cookie-based sessions
- Protocol integration: Model Context Protocol (MCP) over JSON-RPC
- Runtime: Node.js
- Client capabilities: Progressive Web App with service-worker offline fallback
Data model
Core entities include:
- Users and signed sessions
- Thought cards and concept tags
- Daily tasks and recurring task templates
- Day notes and daily check-ins
- Source library items and source ideas
- Insight links between thoughts and ideas
- Behavioural activation modules and entries
- Worry-postponement modules, evidence, postponed items, and experiment days
- Conversation summaries
- RAG documents and embedding chunks
- Per-user AI and retrieval settings
Getting started
Prerequisites
- Node.js
- PostgreSQL with the
pgvector extension enabled
- A Google Gemini API key for AI features
Install and run
npm install
npm run dev
Open http://localhost:3000.
Environment variables
Create .env.local with the database, authentication, and AI settings required by your deployment:
DATABASE_URL=postgresql://user:password@localhost:5432/thoughts
AUTH_SECRET=replace-with-a-long-random-secret
GEMINI_API_KEY=your-gemini-api-key
# Optional Gemini configuration
GEMINI_EMBEDDING_MODEL=gemini-embedding-001
GEMINI_LLM_MODEL=gemini-3.6-flash
EMBEDDING_DIM=1536
# Optional RAG sync lease duration (seconds, default 600; max 3600)
RAG_SYNC_LEASE_SECONDS=600
# Required only for external MCP access
MCP_API_KEY=replace-with-a-secret-api-key
MCP_USER_ID=1
# Optional password-reset email provider (Resend)
APP_BASE_URL=http://localhost:3000
RESEND_API_KEY=your-resend-api-key
RESET_EMAIL_FROM=Thoughts <[email protected]>
# Optional: use a dedicated reset-token hashing secret
RESET_TOKEN_SECRET=replace-with-a-long-random-secret
GOOGLE_API_KEY can be used as an alternative to GEMINI_API_KEY. Never commit real credentials or expose server-only secrets in client code. If the browser-based task agent is enabled, configure its public client key according to the deployment's security model.
For local development, if RESEND_API_KEY is not set, password reset links are logged server-side in non-production mode so you can complete the reset flow without a live email provider.
Development commands
npm run dev
npm run build
npm run start
npm run lint
npm test
Export shape
Daily and monthly exports preserve timestamps, mood scores, task states, check-ins, day notes, linked ideas, and grouped summaries in machine-readable formats. A simplified daily report looks like:
{
"date": "2026-06-22",
"thoughts": [],
"tasks": [],
"task_progress": {
"total": 3,
"done": 2,
"completion_rate": 66.7
},
"day_note": {
"intention": "...",
"note": "..."
},
"check_ins": []
}
Privacy and security
Thoughts is designed for one private user account. Dashboard routes require a signed session, database queries are scoped by user ID, and the MCP endpoint requires an API key plus an explicitly configured user. Use HTTPS in production, keep AUTH_SECRET, DATABASE_URL, GEMINI_API_KEY, and MCP_API_KEY server-side, and treat exported backups as sensitive personal data.
In one sentence
Thoughts is a private AI-assisted personal operating system that turns journaling, tasks, routines, therapeutic exercises, ideas, conversations, and semantic memory into one searchable and actionable workspace.
