IndividuateAI

The "Organic Integral" AI Therapist interface.

Project Setup

This project is built with Rust (Leptos) and Tailwind CSS.

Prerequisites

1. Rust: Install Rust.
2. cargo-leptos: Install the build tool.

   cargo install cargo-leptos

3. Tailwind CSS: Installed locally via NPM.

   npm install

Running the App

To start the development server with hot-reloading:

cargo leptos watch

Open http://localhost:3008 in your browser.

Deterministic privacy E2E test

The standalone harness starts the app with an isolated SQLCipher database, uses a local deterministic OpenAI-compatible stub for chat and embeddings, and drives a Chromium virtual PRF authenticator. It does not read .env or contact an LLM provider.

Requirements: Node.js/npm, Cargo, and Chromium or Chrome.

npm ci
E2E_BROWSER_PATH=/usr/bin/google-chrome npm run test:e2e

The test covers passkey registration/login, chat persistence, logout isolation, process-restart persistence, and database assertions proving that content and vector memory are encrypted and the canary is absent from the database files. The temporary database is written under target/e2e/.

Coolify deployment

GitHub Actions publishes the image to the public GitHub Container Registry:

ghcr.io/dunctk/individuate-ai:latest

In Coolify, configure the application as a public Docker image deployment, using that image and container port 3008. Mount persistent storage at /app/data, and provide the required runtime secrets (COOKIE_SECRET, MEMORY_DB_KEY, OPENAI_API_KEY, and OPENROUTER_API_KEY).

Coolify's source auto-deploy is disabled so it does not deploy before the image exists. Add the Coolify deploy-hook URL as the GitHub Actions secret COOLIFY_DEPLOY_HOOK_URL, and add a Coolify API token with Deploy permission as COOLIFY_TOKEN. The workflow sends that token as a Bearer authorization header after both image pushes succeed. Set the GHCR package visibility to Public once in GitHub package settings; subsequent pushes use the repository's GITHUB_TOKEN.

Model and usage configuration

The production defaults use z-ai/glm-5.2 for therapist and drafting replies, openai/gpt-5.4-nano for structured memory extraction, openai/gpt-4o-mini for lightweight session titles, and text-embedding-3-small for vector recall. Override them with OPENROUTER_MODEL, GRAPH_EXTRACTOR_MODEL, GRAPH_EXTRACTOR_FALLBACK_MODEL, GRAPH_EXTRACTOR_MAX_ATTEMPTS, GRAPH_EXTRACTOR_TIMEOUT_SECONDS, RELATIONSHIP_PROFILE_MODEL, SOCIAL_RELATIONSHIP_MODEL, EPISODE_EXTRACTOR_MODEL, SESSION_SUMMARY_MODEL, and EMBEDDING_MODEL.

Focused previous-chat search uses the quantized BGESmallENV15Q FastEmbed model locally; chat text is never sent to an embedding API. Model artifacts use FastEmbed's cache. Set FASTEMBED_CACHE_DIR to relocate it (HF_HOME takes precedence), and set LOCAL_EMBEDDING_THREADS to cap ONNX inference threads (default 2, clamped to 1–4). The Docker image defaults the cache to /app/data/fastembed-cache, so the existing /app/data persistent mount retains the download. The first semantic search or newly saved user message may download the model; if initialization or inference is unavailable, search continues with lexical ranking. Initialization waits at most 180 seconds and inference waits at most 20 seconds by default; override these with LOCAL_EMBEDDING_INIT_TIMEOUT_SECONDS (clamped to 10–900) and LOCAL_EMBEDDING_INFERENCE_TIMEOUT_SECONDS (clamped to 1–120). Inference is single-permit, so timed-out or concurrent calls cannot accumulate blocking jobs. Hybrid ranking gives user-authored messages a 69% semantic, 28% lexical, 2% recency, and 1% user-evidence score; semantic-only results must meet a conservative 0.55 cosine floor, while lexical matches remain eligible below it.

Session titles are refreshed after the first exchange and every four exchanges thereafter. SESSION_SUMMARY_EVERY_N_EXCHANGES changes that interval. Model prompts retain the most recent 24 messages by default; use MAX_AGENT_HISTORY_MESSAGES to tune the bounded history window. Complete chat history remains encrypted in SQLite and available to focused recall.

Stripe subscriptions

The app is paid-only. Stripe Checkout offers one complete plan at $24.99 USD monthly or $239 USD yearly, and €29.99 monthly or €289 yearly. The EUR prices are configured as tax-inclusive. Plan copy avoids fixed usage claims; the server applies private, environment-configurable capacity safeguards.

Create or verify the product, four recurring prices, tax code, and customer portal configuration in each Stripe mode:

./scripts/setup_stripe_catalog.sh sandbox
./scripts/setup_stripe_catalog.sh live

The script uses stable Stripe lookup keys, so it is safe to rerun and the app does not need price IDs in its environment. Configure these runtime variables:

BILLING_ENABLED=true
ADMIN_EMAIL=d@uncan.net
STRIPE_MODE=live
APP_BASE_URL=https://individuateai.com
STRIPE_AUTOMATIC_TAX=true
STRIPE_SANDBOX_SECRET_KEY=sk_test_...
STRIPE_SANDBOX_PUBLISHABLE_KEY=pk_test_...
STRIPE_SANDBOX_WEBHOOK_SECRET=whsec_... # optional only during local sandbox setup
STRIPE_LIVE_SECRET_KEY=sk_live_...
STRIPE_LIVE_PUBLISHABLE_KEY=pk_live_...
STRIPE_LIVE_WEBHOOK_SECRET=whsec_...

ADMIN_EMAIL is matched case-insensitively against the signed-in account. That account bypasses billing and can open /admin to grant or revoke lifetime complimentary access for registered users. Grants are stored in SQLite with an append-only grant/revocation event log; they do not cancel an existing Stripe subscription.

Create each webhook and immediately copy the one-time signing-secret output into the matching runtime variable:

./scripts/setup_stripe_webhook.sh sandbox https://individuateai.com
./scripts/setup_stripe_webhook.sh live https://individuateai.com

The endpoint subscribes only to Checkout completion and subscription lifecycle events. Finish the applicable Stripe Tax registrations before enabling automatic tax in live mode.

Internal safeguards can be adjusted with MONTHLY_CHAT_SOFT_LIMIT, MONTHLY_VOICE_TOKEN_SOFT_LIMIT, and MONTHLY_TTS_CHARACTER_SOFT_LIMIT. These are operational controls, not advertised plan quotas.

Design System

• Colors: Deep Void Green, Parchment, Integral Turquoise, Systemic Yellow.
• Fonts: Fraunces (Headings), Urbanist (Body).
• Vibe: Botanical, Jungian, Glassmorphism.

Architecture

• Frontend: Leptos (WASM).
• Server: Axum (SSR).
• Styling: Tailwind CSS (v3).

License

Copyright (C) 2026 Duncan.

This project is licensed under the GNU Affero General Public License v3.0 or later (AGPL-3.0-or-later). See LICENSE for the complete terms. If you run a modified version of this software as a network service, you must make the corresponding source available to its users.