Prompt Inspector

Spec sheet

What it does

Prompt Inspector places a real-time security layer between untrusted input and an LLM application or agent.Read 15 Oct 2026 · docs.promptinspector.io

Purpose

Prompt Inspector provides real-time prompt injection detection, sensitive word filtering, and AI safety guardrails through an API call.Read 4 Oct 2026 · promptinspector.io

Integrations

The maker lists a REST API, Python SDK, Node.js SDK, MCP server, and agent skills for OpenClaw and Claude Code.Read 4 Oct 2026 · promptinspector.io

Support

The homepage lists community support for self-hosted users and dedicated SLA support for enterprise customers.Read 4 Oct 2026 · promptinspector.io

License

The core repository and documentation identify the open-source edition as AGPL-3.0 licensed.Read 15 Oct 2026 · docs.promptinspector.io

Agent skills

The integration repository includes zero-dependency Python and Node scripts for OpenClaw, Claude Code and other skill-capable agents.Read 15 Oct 2026 · github.com

AI review

Ambiguous inputs can be escalated to DeepSeek, Qwen or Google Gemini for an additional verdict.Read 15 Oct 2026 · github.com

API limits

The maker's API article lists 10,000 monthly requests and 10 requests per second for Free, while the homepage lists 1,000 monthly detections.Read 4 Oct 2026 · promptinspector.io

Authentication

SDKs accept an API key directly or through the PMTINSP_API_KEY environment variable, while MCP accepts X-App-Key or Bearer credentials.Read 15 Oct 2026 · github.com

Custom filtering

Each tenant can define its own sensitive words and regular-expression rules in addition to the built-in threat library.Read 15 Oct 2026 · docs.promptinspector.io

Detection funnel

Its pipeline combines SHA-256 caching, Aho-Corasick keyword and regex checks, vector similarity, optional LLM review and score arbitration.Read 15 Oct 2026 · github.com

Detection layers

The engine uses hash caching, sensitive word matching, semantic vector analysis, optional AI review, and arbitration.Read 4 Oct 2026 · promptinspector.io

Documentation depth

The documentation site covers architecture, REST use, Python and Node SDKs, MCP setup and open-source deployment.Read 15 Oct 2026 · docs.promptinspector.io

Enterprise

The enterprise offering includes private or on-premise deployment, custom integrations, managed payload library updates, and dedicated SLA support.Read 4 Oct 2026 · promptinspector.io

Enterprise edition

Enterprise adds analytics and audit logs, team roles, SSO/SAML, SLA-backed support, onboarding and custom model fine-tuning.Read 15 Oct 2026 · github.com

Free allowance

The homepage states the free plan includes 1,000 detections per month.Read 4 Oct 2026 · promptinspector.io

Independent architecture praise

AppSecSanta praises the five-stage funnel because inexpensive checks handle obvious threats before slower analysis is used.Read 15 Oct 2026 · appsecsanta.com

Independent cost claim

AppSecSanta says the staged design is intended to cost 60–95% less and run 2–7 times faster than LLM-only detection.Read 15 Oct 2026 · appsecsanta.com

Independent integration praise

AppSecSanta highlights REST, Python, Node.js and MCP support as fitting common development and production workflows.Read 15 Oct 2026 · appsecsanta.com

Independent performance claim

AppSecSanta reports typical response times of roughly 10 to 42 milliseconds and near-zero latency for cached threats.Read 15 Oct 2026 · appsecsanta.com

Independent self-hosting fact

AppSecSanta describes self-hosting with Python, FastAPI, PostgreSQL plus pgvector and Redis, and identifies AGPL-3.0 as the license.Read 15 Oct 2026 · appsecsanta.com

Latency claim

The maker says most requests finish in the first two layers in under 3 milliseconds and full semantic analysis takes about 30 milliseconds.Read 15 Oct 2026 · github.com

Limit: AI review cost

The documentation says the LLM review layer is slower and more expensive and is reserved for gray-zone cases.Read 15 Oct 2026 · appsecsanta.com

Limit: domain availability

AppSecSanta reported that promptinspector.io was offline as of April 2026 while the GitHub repository remained available.Read 15 Oct 2026 · appsecsanta.com

Limit: input size

The repository configuration sets a maximum input length of 5,000 characters.Read 15 Oct 2026 · github.com

MCP clients

The maker identifies Cursor, VS Code Copilot, and Claude Desktop as tools that can connect through its MCP server.Read 4 Oct 2026 · promptinspector.io

MCP integration

An MCP server exposes a detect tool over Server-Sent Events and supports VS Code Copilot, Cursor, Claude Desktop and Dify.Read 15 Oct 2026 · github.com

Node.js SDK

The official Node.js package is installed with npm and provides JavaScript and TypeScript clients with asynchronous detection calls.Read 15 Oct 2026 · github.com

Pricing detail

The maker describes cloud pricing as pay-as-you-grow but does not state a price on the opened homepage.Read 4 Oct 2026 · promptinspector.io

Python SDK

The official Python package is installed with pip and exposes a PromptInspector client with detect and close methods.Read 15 Oct 2026 · github.com

Repository activity

The main public repository shows seven commits, two stars and no forks on the opened page.Read 15 Oct 2026 · github.com

REST API

A REST endpoint accepts JSON input with an X-App-Key header and returns a request identifier, safety result, score, categories and latency.Read 15 Oct 2026 · github.com

Self-hosted setup

The repository documents Docker commands for pgvector and Redis, backend startup with Uvicorn and a separate Next.js frontend.Read 15 Oct 2026 · github.com

Self-hosting

The maker offers a self-hosted option for non-commercial use under the AGPL-3.0 license.Read 4 Oct 2026 · promptinspector.io

Self-hosting stack

Self-hosted deployment uses Python 3.11 or newer, PostgreSQL 15 or newer with pgvector, Redis 7 or newer and Node.js 18 or newer.Read 15 Oct 2026 · github.com

Self-iterating library

The system can generate attack variants from confirmed threats and add them to its vector threat database.Read 15 Oct 2026 · github.com

Semantic detection

Vector embeddings and pgvector are used to catch paraphrased or obfuscated attacks that simple keyword matching may miss.Read 15 Oct 2026 · github.com

Support model

The maker directs enterprise prospects to [email protected] for commercial support and custom arrangements.Read 15 Oct 2026 · github.com

Tenant isolation

Applications receive separate API keys, configurations, rate limits and detection logs.Read 15 Oct 2026 · appsecsanta.com

Threat coverage

The product is designed to detect prompt injection, jailbreaks, sensitive-content leaks and system-prompt exposure.Read 15 Oct 2026 · github.com

Threat detection

Its detection engine checks for prompt injections, jailbreaks, and system prompt leaks.Read 4 Oct 2026 · promptinspector.io

Plans and prices

Every Prompt Inspector plan and price →

Where Specifiction places it

• LLM Security Tools#5 of 318.5