User repository on GitHub
Text-to-SQL where the model never writes SQL — typed, calibrated decisions (TypeSafe Jev) + code-assembled queries
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Repos & SDKs
Jevsql — sarathi-aiml
A weekend experiment in text-to-SQL where the model never writes SQL: Jev fills typed slots — intent, table, state, county, number — each with a calibrated confidence, and Python code assembles the answers into a parameterized SELECT.
- Stars
- 4
- Language
- Python
- Reading time
- 2 min
- Published by
- Community
- Author
- sarathi-aiml
- Last verified
- Oct 2026
This is one of several Jev listings for the same project. The canonical entry is JevSQL — EugeneBoondock.
- Languages
- Python
- Capabilities
- Open source
Highlights
- Jev fills typed slots — intent, table, state, county and number — with calibrated confidences, and code assembles the parameterized SELECT.
- In the README demo, the misspelled 'Kollin' resolved to COLLIN at 0.99 confidence and the 10-employee filter at 0.93.
- The demo query's SQL was built in 647ms of model time at an estimated cost of about $0.00025.
- A --build-catalog step reads Snowflake metadata offline — tables, columns and one sample row — with no model involved.
- The whole API surface is Jev's three primitives: Choice for options, Score for ordered scales, Noul for yes/no probabilities.
Watch out
Copy reviewed against the truncated repo README on 2026-10-15; the capture cuts off inside the 'How it works' section, so install steps, licensing and benchmarks are not documented here. The 647ms and $0.00025 figures are the author's own demo numbers.
Questions about this entry
Short answers drawn from the linked project or page.
It is a weekend experiment in text-to-SQL where the model never writes SQL. Jev answers typed questions about the query — which table, which state, which column the number constrains — and Python code assembles those answers into a parameterized SELECT.
By construction: Jev never produces SQL text or any text. It only answers the typed questions the code defines, each with a calibrated confidence, so the query is assembled deterministically from vetted catalog metadata instead of generated.
The README walks through 'how many companies in Kollin county texas with more than 10 employees': intent=count and table=business at 1.00, state=TX at 1.00, county=COLLIN at 0.99, with the misspelling resolved semantically, and the 10-employee filter at 0.93.
The SQL was built in 647ms of model time at an estimated cost of about $0.00025, per the README. Those are the author's own figures for a single demo query, not a benchmark across a workload.
It is an offline stage that reads Snowflake metadata into a catalog of tables, columns and one sample row without involving the model. The captured README shows it as the first step, before typed slots are filled and SQL is assembled.
The captured README documents a Snowflake metadata path and Snowflake-flavored SQL (TRY_TO_DOUBLE, TO_VARCHAR). The capture is truncated, so support for other databases is not documented and should not be assumed.
It is sarathi-aiml's repository, created and last pushed on 2026-09-21 with 4 stars, and the GitHub metadata declares no license. Treat it as a proof-of-concept weekend experiment rather than a maintained product.
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
A TypeSafe/Jev kernel split out of Lean Refactor Arena: it acts as a gate rather than a proof authority, with choice/score/noul projectors, caches and rankers, and Lean candidates that only a consuming Lean/Lake verifier can admit.
Jev Repos & SDKs#community#catalog-import#logicrw
Community
endomorphosis/jevopsPython
endomorphosis/jevopsPythonLangChain is the agent engineering platform: an MIT-licensed Python framework from the LangChain team for building agents and LLM applications by chaining interoperable components with third-party model, tool, and vector-store integrations.
Jev Repos & SDKs#community#catalog-import#cobanov
Community
langchain-ai/langchainPython
langchain-ai/langchainPythonComposio is an MIT-licensed SDK monorepo giving AI agents 1,000+ pre-authenticated toolkits with per-user sessions, authentication, triggers, and a sandbox, and it ships TypeScript and Python adapters including one for TypeSafe (Jev).
Jev Repos & SDKs#community#catalog-import#cobanov
Community
composiohq/composioTypeScript
composiohq/composioTypeScriptBAML is Boundary's Apache-2.0 programming language for agents: TypeScript-looking syntax, a Rust-like type system that persists at runtime, typed errors, green threads, and callable from TypeScript, Python, Go, C#, and Java.
Jev Repos & SDKs#community#catalog-import#cobanov
Community
boundaryml/bamlRust
boundaryml/bamlRustA PHP monorepo that adds typed decision APIs on top of structured output, unified inference, retrieval and an agent SDK: the Polyglot Decision API can call the TypeSafe provider and return probabilities for noul questions, alongside OpenAI, Anthropic, Gemini and Ollama drivers.
Jev Repos & SDKs#community#catalog-import#logicrw
Community
cognesy/instructor-phpPHP
cognesy/instructor-phpPHPA TypeScript agent harness toolkit built on Effect and Effect AI: define schemas, instructions and tools, and the runtime runs the model/tool loop, validates output, and retains accepted work across process restarts using SQLite, PostgreSQL or Cloudflare hosts.
Jev Repos & SDKs#community#catalog-import#logicrw
Community
danieljvdm/effect-agentTypeScript
danieljvdm/effect-agentTypeScriptFrom the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.