AnkiWeb addon 751846187
Enables semantic and keyword search across selected Anki notes, generating cited AI answers through local or cloud providers.
Open on AnkiWeb GitHub Find alternatives
active
| Min Anki | Max Anki | Updated |
|---|---|---|
| 25.09.2 | + | 2026-06-15 |
Loading…
Enables offline AI semantic and keyword hybrid card search on local CPU, with a browser semantic: prefix, a search dialog for previewing and suspending cards, and automatic model download.
Generates local Ollama vector embeddings for Anki cards, stored in ChromaDB, enabling semantic natural-language searches via "vec:" queries that can combine with keyword filters.
Searches PDF or PPTX lecture slides against your Anki collection using local semantic embeddings, moves matched cards into decks, and uses Gemini to generate cloze cards for uncovered concepts.
Enables natural-language semantic card search via Voyage AI embeddings, optional Claude reranking and smart search, plus duplicate finding, card actions, and clinical vignette/guideline review modes.
Adds a local AI panel in Anki's Browser that turns plain-English requests into Anki searches, expands synonyms, and finds matching cards privately.
Grades written Anki answers using an LLM, retrieves supporting material via semantic RAG, offers tutor explanations, analytics, batch assessment, and optional interval suggestions.
Anki Semantic Search is a smart AI search tool for your Anki collection.
Search your cards by meaning, not just exact words. Ask a question, find the most relevant notes, and optionally generate an AI answer grounded in your own cards with source citations.
It is useful for large medical decks, lecture notes, personal knowledge bases, and any collection where normal Anki search feels too narrow.
For a private local setup, run a local model server first, such as Ollama or LM Studio, then enter its local server URL in Settings. For cloud providers, enter the required API key in Settings before testing the connection or creating embeddings.
The answer panel keeps the response tied to source notes. Citation links can highlight the relevant result rows, and selected notes can be opened in the Anki Browser for review or editing.
The add-on supports both local-first and cloud workflows.
Local options include:
Cloud and API options include:
You can use the same provider for answers and embeddings when supported, or configure separate providers for each role.
The settings window is split into focused areas for:
This lets you keep irrelevant fields out of the index, reduce noise from large decks, and adapt the search UI to your workflow.
Privacy depends on the providers you choose.
If you use local providers for both answers and embeddings, note content can stay on your machine. If you use cloud providers, selected note text may be sent to the chosen embedding provider during indexing/search and to the chosen answer provider when generating answers.
API keys are stored in the add-on configuration. Debug logs are designed to redact sensitive values.
Optional features may require extra Python packages:
pip install sentence-transformers anthropic pyspellchecker
On Windows, using an external Python for Cross-Encoder reranking is recommended if PyTorch has trouble loading inside Anki.
This add-on is designed for both medical and general Anki workflows. It works best after you carefully choose the decks, note types, and fields that should be searchable, then rebuild embeddings after major changes to your collection or provider settings.
Developed by Zahi-Essam.
This update adds a new chat-style search workflow, local retrieval memory, image support for vision models, and improved source controls.
Semantic search and AI answers for your Anki collection. The add-on helps you find relevant notes by meaning, then can generate an answer grounded in the matching cards.
AnkiWeb code: 751846187
Optional features may require Python packages listed in requirements-optional.txt.
751846187, or place this folder in Anki2/addons21/.Ctrl+Enter.Search UI features:
Settings use a shared dark theme across tabs, with consistent rows, controls, tables, section headers, and scrollbars. The API and Styling tabs are width-constrained for comfortable reading on wide monitors, while Search & Embeddings and Note Types & Fields remain full-width work surfaces.
Configure answer and embedding providers.
Answer providers:
Embedding providers:
Useful controls:
Controls retrieval behavior and embedding indexing.
Search strategies:
AI-assisted retrieval:
Clinical accuracy tuning:
Re-ranking:
sentence-transformers.Technical diagnostics:
Embedding actions:
Choose what can be searched and embedded.
This tab is the main place to prevent irrelevant cards from entering the search index.
Controls visual and layout preferences.
Semantic search needs embeddings. Click Create/Update Embeddings after:
The add-on stores embedding data under user_files/ and uses status checks to detect stale or incomplete indexing.
If a note is excluded from indexing, use Review Ineligible Notes to inspect why.
When Local Memory is enabled, the add-on stores compact snippets from the final notes selected for an answer. These snippets stay in user_files/agent_memory.db and are used only to help future retrieval planning.
ollama pull llama3.2
ollama pull nomic-embed-text
http://localhost:11434
Use the server's OpenAI-compatible base URL, commonly:
http://localhost:1234/v1
Make sure the server exposes the endpoints needed for your selected role:
Enter API keys in Settings. Typical choices:
Provider detection helps identify some keys automatically. Custom endpoints can be set through the API URL field where supported.
Optional packages are not required for basic search, but enable extra features.
Install into the correct Python environment:
Packages:
pip install sentence-transformers anthropic pyspellchecker
Feature mapping:
sentence-transformers: Cross-Encoder re-ranking.anthropic: Anthropic streaming support.pyspellchecker: spell check and right-click suggestions in the search box.On Windows, using an external Python for sentence-transformers is recommended because torch can fail to load inside Anki's Python.
What may be sent externally depends on your provider choices.
API keys are stored in config.json in the add-on folder. Do not share this file.
Logs and user data are stored under user_files/ where possible. Logs redact sensitive values.
The add-on formats note display text separately from search text so HTML blocks and selected fields remain readable in previews and tooltips. Reopen the search dialog or rerun the search after updating.
MIT
Author: Zahi-Essam
Built for medical and general Anki search workflows with semantic retrieval, local AI support, and cloud provider support.