Holmes AI
Ask Ernest Holmes a question and get an answer in his own voice, sourced from his own writing.
ernestholmes.us · Version 2.0
Ernest Holmes (1887–1960) founded Religious Science and wrote the texts that Centers for Spiritual Living still teach from. Holmes AI puts those writings behind a conversation: you bring a question, and the answer comes back in his 1920s–40s cadence — or in plain modern language, if that lands better.
Built for CSL members, Science of Mind students, and anyone who has picked up The Science of Mind and wanted to ask it something back.
How it works
Holmes AI is not a fine-tuned model. It is Claude Haiku 4.5 steered by two things:
- Persona prompts that fix the voice, the vocabulary (Principle, Infinite Mind, Spiritual Law) and the shape of an answer.
HOLMES_PROMPT for his own cadence, MODERN_PROMPT for contemporary language, chosen per request.
- A quote library drawn from his books, so answers can cite his actual sentences instead of paraphrasing from memory.
The corpus was assembled offline. downloads/ holds the Python scrapers and the processed source texts — 463,148 words across Creative Mind and Success (1919), The Hidden Power of the Bible (1929), The Science of Mind (1938), This Thing Called You (1948) and Words That Heal Today (1949). downloads/training_data/ holds what the running app actually reads: quote, Q&A, treatment and explanation sets extracted from those texts.
That split matters for what the app will and won't do. It serves quotes it has rather than inventing new ones; where it has nothing to cite, it answers in voice without a citation.
What it does
- Two voices — "In His Words" for Holmes' own cadence, "Modern" for plain contemporary language.
- Source citations on quoted material, traceable back to a book.
- Treatment generator — builds a Science of Mind spiritual mind treatment around the question asked.
- Question history with bookmarking, search, filtering and automatic tagging by category.
- Quotes browser over the Holmes quote library.
- Email sharing of an answer or a treatment.
- Admin view for reviewing submitted questions.
- Light and dark themes, keyboard navigable, WCAG AA.
Stack
| Layer | Choice |
|---|
| Framework | SvelteKit 2 on Svelte 5 (runes) |
| Language | TypeScript |
| Styling | Tailwind CSS 4 |
| Model | Claude Haiku 4.5 via @anthropic-ai/sdk |
| Storage | SQLite via better-sqlite3 |
| Build | Vite 8, @sveltejs/adapter-node |
| Tests | Vitest + Testing Library |
| Deploy | Docker (Node 20 alpine) behind nginx |
Running it
Node 20+.
cd holmes-gpt
npm install
cp env.example .env # then add your Anthropic key
npm run dev
The SQLite database is created on first run.
Environment
| Variable | Required | Purpose |
|---|
ANTHROPIC_API_KEY | yes | Model access |
ANTHROPIC_MODEL | no | Override the default model id |
DB_DIR | no | SQLite location |
ADMIN_PASSWORD | no | Enables the /admin view |
SECURITY_TOKEN | no | Bearer token for GET /api/security metrics |
Docker
cd holmes-gpt
docker compose -f docker-compose.local.yml up --build
Serves on :3015, with the database on a named volume so it survives a rebuild.
Checks
npm run check # svelte-check
npm run lint # prettier + eslint
npm run test:run # vitest
Abuse controls
The app calls a paid API from an unauthenticated public page, so the server enforces its own ceilings before a request reaches the model:
| Limit | Value |
|---|
| Requests per 15 minutes | 10 |
| Requests per day | 50 |
| Tokens per day | 50,000 |
| Max message length | 1,000 characters |
Alongside those: user-agent bot detection, a 10/hour cap on email sends, 5 admin login attempts per 10 minutes, constant-time password comparison, and counters for requests, rate-limit hits, bot detections and spend, exposed at GET /api/security behind a bearer token.
Layout
holmes-gpt/
├── src/
│ ├── lib/
│ │ ├── server/ # claude.ts, prompts.ts, knowledge.ts,
│ │ │ # security.ts, gate.ts, classify.ts, db/
│ │ ├── components/ # Svelte UI
│ │ ├── state/ # runes-based stores
│ │ └── data/ # treatment templates
│ └── routes/
│ ├── api/ # chat, quotes, treatment, history,
│ │ # emails, admin, security, health
│ └── … # about, support, privacy, admin
├── downloads/ # corpus + Python processing scripts
└── resources/ # source and provenance notes
docs/ # design and implementation notes
License
Private and proprietary. All rights reserved.
Not affiliated with or endorsed by Centers for Spiritual Living.
