User repository on GitHub
Typed semantic decisions for Unix pipelines and CI, powered by TypeSafe AI Jev.
- Forks
- 10
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Tools & CLIs
SemDecide
A CLI that turns Jev into Unix pipeline primitives: is, choose, score, filter, and guard commands return typed verdicts, stable exit codes, and versioned JSON without prompt or parsing glue.
- Stars
- 80
- Language
- Python
- License
- MIT
- Reading time
- 4 min
- Published by
- Community
- Author
- sharziki
- Last verified
- Sep 2026
About this project
Typed semantic decisions for Unix pipelines and CI, powered by TypeSafe AI Jev. SemDecide is grep for meaning and jq for judgment. Pipe in text or JSONL and get a predicate, route, score, filtered stream, and stable process exit code without writing prompt, parsing, retry, or confidence-handling glue. Exact string rules cannot express concepts such as: “This customer is likely to churn.”. “This change introduces a breaking API behavior.”. “This record may describe account takeover.”.
Quoted from the project's README. Read the full README.
Highlights
- Commands cover semantic predicates (is), routing among named options (choose), ordered rubrics (score), JSONL filtering (filter), and agent safety (guard).
- With --json, results carry a schema_version, probability, threshold, model id, and token usage; failures emit a versioned error object on stderr.
- Input is rejected before Jev if it is empty, binary, invalid UTF-8, oversized, or malformed; defaults are 1,000,000 bytes, a 10 s timeout, and 2 retries.
- The guard recipe checks authorization, destructiveness, ambiguity, secret exposure, and consequence, then local code returns allow, escalate, or block.
- The runtime is dependency-free and installs from the signed-off GitHub release wheel because it is not yet on PyPI.
Quickstart
bash
printf '%s' 'Login from a new country, followed by payout changes.' \
| semdecide is 'This describes a plausible account takeover'Watch out
MIT-licensed and independent, not an official TypeSafe product; needs Python 3.10+ and a TYPESAFE_API_KEY, and the README warns it is not an authorization system, sandbox, or security proof.
Questions about this entry
Short answers drawn from the linked project or page.
A layer, and the README says so plainly: SemDecide is not an authorization system, sandbox, security proof, or tool executor, and semantic decisions can be wrong. It advises keeping deterministic permission checks around money, credentials, production infrastructure, private data, and irreversible operations. The failure direction is at least safe: in guard, provider errors fail closed to escalate rather than allow.
grep matches text; SemDecide evaluates a plain-English predicate against whatever you pipe in and returns a probability plus a verdict, so you can write "this describes a plausible account takeover" instead of a regex. It also refuses to launder uncertainty into a boolean: --threshold and --uncertainty-margin define a band where a borderline answer exits 3 rather than being forced true or false.
The contract is documented as grep-like with uncertainty and provider failure kept distinct: 0 for true, selected, scored, or at least one definite filter match; 1 for false or no matches; 2 for invalid local input or usage; 3 for uncertain under your margin or confidence floor; 4 for provider, auth, timeout, or invalid-response failure. guard uses its own numbers and keeps deprecated check and reflex aliases for one minor release.
It goes to TypeSafe AI, and the README tells you not to submit material your data policy forbids sending there; no local model path is documented. What stays local is the pre-flight: empty, binary, invalid UTF-8, oversized, or malformed input is rejected before any call, with a 1 MB default cap, a 10-second per-attempt timeout, and 2 retries. Credentials come from the environment or a 600-mode credentials.env, never argv.
It needs Python 3.10 or newer and is not yet on PyPI, so install the signed GitHub release wheel directly with pipx install or uv tool install pointing at the v0.2.1 wheel URL. From a source checkout, create a virtual environment and pip install -e . Then set TYPESAFE_API_KEY, or put it in ~/.config/typesafe/credentials.env with mode 600.
is for a semantic predicate, choose to route among named options, score for ordered rubrics, filter to select matching JSONL records, and guard as an opinionated agent-safety recipe returning allow, escalate, or block. is, choose, and score emit stable JSON with --json, and filter adds _semdecide metadata to matching records unless --raw is set.
It rejects empty, binary, invalid UTF-8, oversized, and malformed structured input before calling Jev. --max-input-bytes caps input at 1,000,000 by default, --timeout sets a 10-second per-attempt provider timeout, and --retries defaults to 2 with a maximum of 5; --quiet suppresses output and leaves only the exit code.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.
FacebookClinicnotes
YouTube: Kubernetes admission control with a decision model, fail-closed.
X: OpenRouter Decision Model Rankings
X: herdr + jev, an agent router
More Jev resources like this
A pre-commit hook that answers five Jev questions about the commit message and staged diff together, warning on mismatches and blocking only on a credential-shaped line.
Jev Tools & CLIs#community#python#coding-agent
Community
valentynkit/jev-commitPython
valentynkit/jev-commitPythonA Node CLI that sends source, configuration, build, and CI files to Jev and reports suspicious files and lines with probabilities, as a second opinion before running unfamiliar code.
Jev Tools & CLIs#community#typescript#cli
Community
luantak/is-maliciousTypeScript
luantak/is-maliciousTypeScriptPython MCP server that exposes classify, score, check, match, and screen tools to MCP-compatible agents, returning typed judgments over a closed answer set with token usage and latency. Ships generic triage and routing question examples.
Jev Tools & CLIs#community#python#mcp
Community
blakestone-x/jev-mcpPython
blakestone-x/jev-mcpPythonAgent-ergonomic CLI following the AXI conventions that gives coding agents Jev judgments for blocking risky tool calls, screening fetched content for prompt injection, triaging logs, flagging risky diffs, and ranking many items. Routine commands are decided locally at no cost.
Jev Tools & CLIs#community#typescript#cli
Community
shiftynick/jev-axiTypeScript
shiftynick/jev-axiTypeScriptA Hermes Agent plugin that replaces the auxiliary approval reviewer with TypeSafe's Jev decision model: one request asks six typed questions and deterministic code returns APPROVE, DENY, or ESCALATE.
Jev Tools & CLIs#community#python#agents
Community
anpicasso/hermes-jev-approvalsPython
anpicasso/hermes-jev-approvalsPythonA safety guardrail for coding agents (Claude Code, Codex, Cursor, Antigravity, Grok) that intercepts shell commands, script executions and file writes before they run. Its neural tier can use Liquid AI d1-3B, CLM, TypeSafe Jev cloud or pure rules; a deterministic tier blocks destructive patterns in under 0.1 ms.
Jev Tools & CLIs#community#github#security
Community
Bl4ckk01/agent-guardrailPython
Bl4ckk01/agent-guardrailPythonFrom 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.