User repository on GitHub
Semantic SQL for Postgres, powered by Jev
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
Jev directory
CommunityJev Tools & CLIs
jevQL
psql-shaped CLI and Go, TypeScript, and Python SDKs that let you write jev(), jev_prob, jev_choice, and jev_score in queries against vanilla PostgreSQL, judging the surviving rows with Jev in batches. No CREATE EXTENSION or superuser required.
- Stars
- 14
- Language
- Go
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- kylemclaren
- Last verified
- Sep 2026
About this project
Semantic SQL for vanilla PostgreSQL. One extra family of functions, jev(), works in any query, from the CLI, from a shared HTTP and MCP node, or from the Go, TypeScript and Python SDKs. The database only ever sees ordinary SQL. jev calls are evaluated by the CLI using TypeSafe's System One model (Jev). No CREATE EXTENSION, no superuser, no wire-protocol proxy. Prebuilt binaries for macOS and Linux (arm64 and amd64) are attached to each GitHub release as jevql .tar.gz with a checksums.txt.
Quoted from the project's README. Read the full README.
Works forJev for Tool calling
Used inJev for Data & Analytics
Highlights
- Adds jev(), jev_prob, jev_choice, and jev_score to ordinary SQL on vanilla PostgreSQL with no extension.
- Postgres runs the plain SQL; Jev judges the surviving rows in batches and answers are cached in local SQLite.
- Statements without jev_* pass straight through, and --explain shows the row count and cost estimate before spending.
- Ships a psql-shaped CLI plus Go, TypeScript, and Python SDKs and a shared HTTP and MCP node.
- Every row that survives the SQL filters is sent to TypeSafe, so cheap predicates belong in SQL first.
Quickstart
bash
brew install kylemclaren/tap/jevql
export TYPESAFE_API_KEY=YOUR_API_KEY
export DATABASE_URL=postgres://user:pass@localhost:5432/app
jevql -c "SELECT * FROM people WHERE jev(people, 'could work from home')"Watch out
MIT-licensed. Needs a Postgres connection and a TYPESAFE_API_KEY; building from source requires Go 1.23+ with CGO and a C compiler because libpg_query is bundled.
Questions about this entry
Short answers drawn from the linked project or page.
No. There is no CREATE EXTENSION, no superuser, and no wire-protocol proxy. The database only ever sees ordinary SQL, because the CLI parses the statement, strips the jev_* calls, judges the surviving rows outside Postgres, and then projects booleans, probabilities, choices, and scores back in process.
Version one supports a single SELECT, with a WITH allowed only if the CTE contains no jev_* calls. It errors rather than guessing on jev_* in INSERT, UPDATE, DELETE, or DDL, in a CTE or subquery, in HAVING, under a window function, with OR, on comparisons like jev_prob(...) > 0.7, on SELECT DISTINCT, on set operations, and on tables with bytea columns.
Every column of the judged alias, unless you narrow it. jev(alias, 'question') packs all of that relation's columns into a JSON row object and sends it over HTTPS to TypeSafe, while the column-list form or --columns sends less. The README is blunt: do not use this on data you cannot share with TypeSafe.
Yes. --explain prints the collect SQL, the row count after SQL filters, the batch count, a rough token estimate, and a cost, while making no TypeSafe requests. --max-rows defaults to 2500 and aborts before any HTTP call if collection exceeds it, and --max-chars can cap row-object size.
The jev() family: jev and jev_prob for boolean or probability, jev_choice for a label, jev_score and jev_score_norm for weighted levels, jev_confidence for the provider confidence, and jev_eval for the raw answer JSON.
Answers are cached in local SQLite keyed by sha256 of model, kind, question, options and canonical row JSON; there is no expiry in v1, so clear it with the cache command or bypass it with --no-cache.
Row objects are canonicalised and deduplicated, then misses are packed --batch-size rows per request per question and sent with --concurrency workers; Ctrl-C cancels in-flight work.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: jev(): Postgres WHERE clauses in plain language
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 NewsMore Jev resources like this
A PostgreSQL extension that filters, ranks, and classifies table rows with plain-language conditions judged by Jev, batching 20 rows per request and caching answers per session.
Jev Tools & CLIs#community#postgres#sql
Community
realZachi/pg-jevShell
realZachi/pg-jevShellAn open-source, commercially hosted generative-engine-optimization platform that uses Jev judgments within a broader application to track brand mentions and placement in AI answers across ChatGPT, Claude, Gemini, and Perplexity.
Jev Tools & CLIs#community#typescript#postgres
Community
usenotra/notraTypeScript
usenotra/notraTypeScriptA single-binary Go MCP server exposing one evaluate tool for noul, choice, and score questions, with one-command registration for Claude Code, Claude Desktop, and Codex.
Jev Tools & CLIs#community#go#mcp
Community
itsmostafa/typesafe-mcpGo
itsmostafa/typesafe-mcpGoA pre-alpha PostgreSQL extension for categorical work in SQL: Noul, Choice, and Score functions plus batched variants that classified 38 strings in two requests.
Jev Tools & CLIs#community#postgres#sql
Community
giuliosmall/pg_typesafeC
giuliosmall/pg_typesafeCRust coding-agent harness (MIT, ~20k stars) whose crates/jcode-base/src/jev.rs is a typed TypeSafe Jev decision transport, used for memory pruning (memory_jev.rs), browser navigation (browser_jev.rs), and voice routing (voice_intent.rs) with separate provider selectors per purpose.
Jev Tools & CLIs#community#rust#coding-agent
Community
1jehuang/jcodeRust
1jehuang/jcodeRustVercel Labs' Unix-like coding agent CLI written in Zig, shipping a typesafe_permission_reviewer builtin that runs the agent's permission decisions through Jev rather than an LLM call. Model-agnostic and embeddable via libfx and ACP.
Jev Tools & CLIs#community#coding-agents#cli
Community
vercel-labs/fxZig
vercel-labs/fxZigFrom 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.