Quran, hadith and names of Allah in one app and API.
Reminder brings daily reminders, the Quran, hadith and the Names of Allah together in one app and API. Reading, listening and returning to these texts remain the core experience.
Search helps readers find passages; AI can add source-backed summaries and explanations. Arabic language resources, translations and tafsir support deeper understanding of the text. They should enrich reading without creating a separate general chat or broad knowledge destination. See project purpose.
- Quran in English & Arabic
- Quran Audio Recitation (Arabic & English)
- Names of Allah & Meaning
- Sahih Bukhari and Sahih Muslim in Arabic and English
- Index & Search with RAG and LLMs
- Optional Claude, Ollama or OpenAI integration
- API to query Quran, hadith, names
- Daily reminder web notifications
Find the latest release
Or Go install
go install github.com/asim/reminder@latest
Quick Start (reading and SQLite search need no model API key):
reminder --serve --webSearch refreshes the bundled Quran, tafsir, Names, Bukhari and Muslim sources in
~/.reminder/search.db at startup. Collection updates do not require embeddings.
LLM Configuration (optional) - used for answer summaries:
# OpenAI (default: gpt-6-luna)
export OPENAI_API_KEY=xxx
export OPENAI_MODEL=gpt-6-luna # optional override
# Use a specific Ollama model (default: gemma3:1b)
export OLLAMA_LLM_MODEL=gemma3:1b
# Use Claude when OpenAI and an explicit Ollama model are not configured
export ANTHROPIC_API_KEY=xxxAn explicit Ollama model takes priority, then OpenAI, then Anthropic. OpenAI API credentials are required; a ChatGPT/Codex subscription is not an API key.
The reader links back to the full chapter at the selected verse.
Search offers All, Quran, Hadith, Tafsir and Names filters, applied before result
limits. /api/search accepts source: "all" | "quran" | "hadith" | "tafsir" | "names".
The browser retains both searches and AI answers in Recent, separately by source
and mode; refresh restores the query and filter without regenerating a saved answer.
Ask AI and verse explanations use native tool calling with OpenAI-compatible (including tool-capable Ollama) and Anthropic models. The model can answer directly or repeatedly inspect lookup results and refine its investigation. Available tools search the texts, retrieve a verse and its neighbours, scan Quran mentions, and look up Arabic dictionary/morphology evidence. There is no mandatory planning call.
The prompt allows language knowledge, context and reasoning to support explanation. Quotations, references, counts, exhaustive lists and attributed scholarly claims must be verified against retrieved sources. Source numbers remain stable across all tool rounds. Mention inventories count explicit whole-word matches for supplied names/aliases, not thematic or unnamed references. General verse explanations start with the passage's meaning, supported by context and tafsir where useful.
Research is bounded by a 120-second deadline, 1 MiB of conservative cumulative request/output allowance, and an eight-turn loop backstop (the last turn is answer-only). Existing per-client, key and global usage limits apply to every provider call, including tool schemas and results. Local models must support native tools; provider errors are surfaced rather than silently switching to an unverified answer. No external web browsing is enabled.
Summaries are optional. With no model configured the summarise controls are hidden and search still returns references.
Optional semantic search supplement:
export OPENAI_API_KEY=xxx
# Enabled automatically; set SEARCH_SEMANTIC=false to disableThis combines the existing vector and current SQLite rankings using reciprocal
rank fusion. It is enabled automatically when an OpenAI key is configured. SQLite coverage is retained if embeddings are unavailable;
new collections do not need a vector rebuild to appear in search or model evidence.
--index, --export and --import remain explicit vector maintenance commands.
See hadith collections for source numbering and collection URLs.
Run the server
reminder --serve
Go to localhost:8080
All queries are returned as JSON
/api/latest- for the latest reminder with LLM-generated contextual message- Returns: verse, hadith, name of Allah, links, and an AI-generated spiritual message
- The
messagefield contains an LLM-generated reflection (2-3 sentences) based on the verse, hadith, and name - Falls back to default message ("In the Name of Allah—the Most Beneficent, Most Merciful") if LLM is unavailable
/api/quran- to get the entire quran/api/names- to get the list of names/api/hadith- to get the entire hadith/api/search- to get summarised answerqparam for the queryPOSTusingcontent-typeasapplication/jsoncurl -d '{"q": "what is islam"}' http://localhost:8080/api/search
See /api for more details
MCP clients can connect to https://reminder.dev/mcp using HTTP POST for
Quran, hadith, names, reminders and search tools. Search summaries are opt-in
with summarise: true (default: false).
The reminder bakes in a "lite" app by default. This can be replaced by a featureful react app.
To build the react app
# requires Node.js 22 and npm
make setup
Build the app
make build
Pass the --web flag to use the react app
reminder --web --serve
The Quran says in 6:90
Say,
“I ask no reward of you for this (Quran) —
it is a reminder to the whole world.”
See the Arabic source audit for verification scope and limitations.
The sources used for data
File an issue to discuss
Quran, Bukhari, Muslim and names are saved for offline reading after the initial download.
Audio, search and AI require a connection. Use make build to include the service worker.
POST /api/model/ask requires authorisation: Authorization: Bearer <key>.
Operators issue keys on the CLI; see model access and usage.
Summaries require summarise: true. Public search defaults: 100 requests/day,
20 per IP, two concurrent generations and 1,024 output tokens. Configure with
AI_DAILY_LIMIT, AI_IP_DAILY_LIMIT, AI_CONCURRENCY_LIMIT and AI_MAX_OUTPUT_TOKENS.
Daily/hourly generation is exempt. Counters persist across restarts for a single server.
Nginx's X-Real-IP is trusted from loopback; set TRUSTED_PROXY_CIDRS for other proxy addresses.