Discover the Best AI Tools

Jev Switch routes typed Jev requests through configurable provider graphs using a Rust daemon, with web and desktop controls. Upstream services perform inference; local mode is not necessarily offline, and the public interface is Jev rather than a general Chat Completions API.

This independent OpenClaw plugin exposes an explicitly invoked TypeSafe decision tool without background hooks or a chat-model provider. Compatibility evidence is limited to OpenClaw 2026.9.4 on the documented environment; it sends supplied state externally and conflicts with another plugin using the same ID.

pytest-jev adds probability-based semantic assertions to pytest and batches claims about one output into a Jev request. It requires Python 3.10+, pytest 7.4+ and an API key; without a key the affected tests are skipped, and model judgments are not deterministic truth.

Ground Zero evaluates response grounding, correctness against expected answers and instruction following using Jev judgments. It is an alpha library requiring OpenRouter access; schemas may change and inconsistent judge outputs need manual validation.

Hiep Paseo plugins supplies independent local panels and tools for conversation boards, workspaces, task context and optional Jev delegation. Its Jev orchestrator needs Paseo and Node.js 22+; selected task text reaches Jev, consumes quota and routing choices are heuristic.

n8n-nodes-typesafe-ai adds Jev yes/no, choice and score operations to n8n workflows. It requires TypeSafe credentials and sends supplied data to the API; the unverified community package is for self-hosted n8n until cloud verification is available.

Chat2Jev converts Chat Completions requests into editable Jev state/questions and compares results in a local workbench. It requires Node.js 22 and provider credentials; it has no built-in authentication or streaming, and remembered keys are plaintext in browser storage by default.

hunch for TypeScript wraps Jev judgments as array-preserving classification, filtering, ranking and verification functions with probabilities and caching. It needs Node.js 20+ and a TypeSafe key; each distinct item can incur an API call and all object fields are shared unless columns are restricted.

Signalint runs and caches JavaScript/TypeScript diagnostics, clusters repeated issues and tracks recurring diagnostic loops through MCP. It requires a supported Node.js version and project tools; TypeScript checks need tsconfig.json and other languages are unsupported.

Tokenectomy Razor filters framework noise from error traces and redacts recognizable credentials through a local Rust gateway and MCP server. It also offers caching and token limits; filtering and redaction must not be treated as complete leak prevention or a fix for provider quotas.

Token Savior combines structural code navigation, persistent memory and Bash-command rewriting for coding agents. It requires Python 3.11+ and an MCP client; headline performance claims use an unpublished benchmark and remain independently unverified.

x402-mcp discovers paid APIs from CDP Bazaar and exposes them to agents while handling x402 USDC payments. It requires a funded wallet and signing credentials; the configurable per-call price cap is not a total spending budget.

Writing Style Checker flags passive voice, vague wording, repetition and patterns common in AI-generated prose through an editor, API or CLI/MCP. Its explainable pattern checks suggest edits; they cannot prove whether a person or AI authored the text.

replicant-mcp lets coding assistants build Android packages, operate emulators, inspect logs and interact with app interfaces. It requires Node.js 18+ and the Android development toolchain; device-changing actions include uninstalling apps and clearing data.

devpulse provides coding assistants with a local diagnostic snapshot of Git state, running development services, ports and recent errors. It needs access to the project environment; detected issues and redacted logs still require review before acting or sharing.

Tactual models screen-reader navigation cost from browser snapshots and keyboard probes, with CLI, CI and MCP access. It requires Node.js 20+; simulated assistive-technology profiles complement, rather than replace, real screen-reader and manual validation.

prompt-to-asset creates and packages icons, logos and other visual assets through CLI/MCP, using inline SVG, external prompts or provider APIs. It requires Node.js 20.11+; free or keyless routes depend on provider availability and are not guaranteed unlimited.

Markovo converts supported files and explicitly supplied public HTTPS pages into structured Markdown through a CLI or MCP client. Conversion runs on its account-metered service, not locally; credentials and an explicit credit ceiling are required.

TypeSafe Conversation is an unofficial Home Assistant integration that maps Assist commands to built-in device intents using Jev. It sends commands and exposed entities to TypeSafe with your API key; uncertain, conversational or free-text requests can use a configured fallback. It requires Home Assistant 2025.6 or later and currently uses English confirmation templates.

judging-with-typesafe includes Python tools for typed Jev queries and paragraph-level text review using criteria and uncertainty bands. It sends selected paragraphs to TypeSafe and writes scores to JSON, requiring an API key. It can help triage claims for human review, but cannot verify facts or sources; thresholds need evaluation for the material.

jev-layer exposes bounded capability and model recommendations, supervision and decision receipts for agent hosts through CLI and MCP integrations. Hosts retain permission checks and execution. The default provider is an offline demo; real providers need configuration. Optional features fail back to the host, and the experimental browser path is neither a browser executor nor a security boundary.

Tripwire is AI SDK middleware and an OpenAI-compatible proxy for Jev-based checks on response text. Policies can flag or replace outputs, but default passthrough streaming judges after users see them, and judge failures fail open. Tool-only turns are uninspected. It requires service credentials, and the project has not established accuracy against real Jev.

jev-ui provides React components that use Jev judgments to select existing views, rank candidates and show optional interface elements. Developers retain thresholds, fallbacks and actions. Live decisions require server-side TypeSafe credentials and send supplied state to the service; replay or mock mode is not live inference, and presentation gates are not authorization controls.

Moth analyzes supplied error output, redacts likely secrets and retrieves project-local fix history to produce a structured debugging brief. It records passed or failed outcomes reported after verification. It does not crawl repositories, run commands, edit code or perform live research itself; callers remain responsible for executing and truthfully reporting verification.