User repository on GitHub
Batched natural-language judgments for SQLite, powered by TypeSafe Jev
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Integrations & Adapters
Sqlite Jev
Loadable SQLite C extension and Python package that exposes Jev Noul, Choice, and Score judgments as SQL scalar functions and a jev_rows virtual table which batches up to 40 rows into one shared state and joins typed results back to the source by rowid.
- Stars
- 4
- Language
- C
- Published by
- Community
- Last verified
- Oct 2026
About this project
Natural-language predicates, classification, and scoring for SQLite, powered by TypeSafe Jev. It is inspired by pg-jev, but shaped around SQLite's loadable-extension and virtual-table APIs. jevrows reads the selected columns, puts up to 40 rows in one shared state, asks one question per row, and returns a virtual table that can be joined to the source by rowid. Results are cached for the lifetime of the SQLite connection. SQLite 3.45 or newer, built with JSON support and loadable extensions.
Quoted from the project's README. Read the full README.
Used inJev for Data & Analytics
- Capabilities
- Runs locallyRequires API key
Highlights
- jev_rows batches up to 40 source rows per request and returns source_rowid, answer, probability, choice, score, and confidence.
- Requires SQLite 3.45+ with JSON and loadable extensions, a C11 compiler, libcurl, and the TYPESAFE_API_KEY environment variable.
- Connection-local settings cover model, batch_size (40), max_rows (500, a spend guard), timeout (90s), and api_url.
- The sqlite-jev Python package has no runtime Python dependencies and supports Python 3.10+ via sqlite_jev.load().
- A deterministic mock suite runs without calling TypeSafe; a live integration test runs every two months in GitHub Actions.
Watch out
No license is declared in the repository (GitHub reports none); row contents are sent to TypeSafe, so this is a semantic full scan that should be preceded by deterministic SQLite filters.
Questions about this entry
Short answers drawn from the linked project or page.
The jev_rows virtual table puts up to 40 rows into one shared state and asks one question per row in a batched request; scalar functions are for single records, and scan results are cached for the connection lifetime.
Yes. The sqlite-jev package ships the same native extension with no runtime Python dependencies; call sqlite_jev.load() on Python 3.10+ with loadable-extension support enabled in the interpreter's sqlite3 module.
SQLite 3.45 or newer built with JSON support and loadable extensions, a C11 compiler with make, the libcurl runtime as libcurl.so.4 on Linux or libcurl.4.dylib on macOS, and a TypeSafe key. Then run make, export TYPESAFE_API_KEY, and inside SQLite issue .load ./build/jev followed by SELECT jev_version() to confirm. That environment variable is the only one used for the key.
The primitive is the third argument of jev_rows: noul for a boolean judgment, choice with a JSON object of candidate keys and their criteria, or score with a JSON array of ordered levels. Every shape returns source_rowid, answer, probability, choice, score and confidence, with NULL for fields the chosen primitive does not use. Keep the cut-off in SQL rather than in the question so it can rise with the cost of a false positive.
Settings are connection-local through jev_config: model, batch_size at 40, max_rows at 500, timeout at 90 seconds and api_url. max_rows is a spend guard that fails a jev_rows scan before sending any data. The selected columns of every row you scan do go to the model endpoint, so prefilter into a temporary table first; jev_stats() reports connection-local usage, jev_cache_clear() clears the cache, and jev_config never returns the key, only set or unset.
It needs a rowid table. To keep a scan small and cheap, materialize the filtered rows into a temporary table and evaluate that one, and pass only the columns the judgment actually needs, since both accuracy and cost depend on what you include. Results come back keyed by source_rowid, so you join them to the source table in ordinary SQL.
make test runs a deterministic suite against a local mock server and never calls TypeSafe, while make live-test runs the ticket-triage demo, which makes two live requests. CI covers mock tests plus wheel build and test on Linux x86_64 and arm64 and macOS on Intel and Apple Silicon, with a scheduled live smoke test every two months, and release archives with platform Python wheels are published. No license file is declared in the repository, so there are no granted rights to rely on.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: A MySQL plugin that filters rows by asking Jev whether a column matches a sentence.
maayanlevy-hn2 pointsA MySQL plugin that filters rows by meaning (built on TypeSafe Jev)Read the thread on Hacker NewsX: jev(): Postgres WHERE clauses in plain language
More Jev resources like this
4GitHub stars
A MySQL plugin that filters rows with a natural-language condition. AILIKE asks Jev whether a column value matches, and returns 1, 0, or NULL.
Jev Integrations & Adapters#community#sql#database
Community
maayanlevy/mysql-ailikeC++
maayanlevy/mysql-ailikeC++3GitHub stars
Alpha playground that answers plain-English questions over SQLite by composing small Jev decisions -- table, column, comparison and operand selection -- into inspectable typed plans, deterministic SQL and previewed guarded writes. Bundles Cyberchase, DESE and Moneyball databases.
Jev Integrations & Adapters#community#sqlite#text-to-sql
Community
Amine-LG/IntentSQLPython
Amine-LG/IntentSQLPython10GitHub stars
Local-first knowledge search with seven selectable pipelines: BM25 plus Jev reranking is the default vector-free path, plus agentic, embedding-hybrid, taxonomy, gate and line-search modes. Reranks up to 30 candidates in one batched set of Jev Noul relevance questions and streams cited answers from a configurable model.
Jev Integrations & Adapters#community#rag#retrieval
Community
aifabrice/jev-ragPython
aifabrice/jev-ragPython10GitHub stars
A local Jev-compatible server that reads Choice, Score and Noul answers from the logits of any MLX checkpoint or OpenAI-compatible endpoint with no training, serving POST /v1/systemone as a drop-in for the official TypeSafe SDK.
Jev Models & Reimplementations#community#mlx#calibration
Community
bradAGI/rulingPython
bradAGI/rulingPython39.6kGitHub stars
Volcengine's agent context database that unifies memory, knowledge RAG and skills under a viking:// virtual filesystem; it ships a Jev rerank client that scores candidate documents against api.typesafe.ai and treats the probability as relevance.
Jev Integrations & Adapters#community#agent-memory#rag
Community
volcengine/OpenVikingPython
volcengine/OpenVikingPython8.2kGitHub stars
A semantic cache for LLM queries, fully integrated with LangChain and llama_index. Zilliz's cache integrates TypeSafe Jev Noul checks to evaluate cache hit freshness and time-dependent query validity, cutting repeated requests and API cost.
Jev Integrations & Adapters#community#semantic-cache#langchain
Community
zilliztech/GPTCachePython
zilliztech/GPTCachePythonFrom the Jev community
Posts from builders shipping with Jev right now.