Prompt Inspector: AI Prompt Security Review
promptinspector.io · checked 15 Oct 2026
Prompt security that screens AI inputs for attacks, leaks, and sensitive content.
How it plugs in
Signal on LLM Security Tools7.7#1 of 31
Prompt Inspector places a live security layer between untrusted text and an AI application or agent. Its API checks incoming prompts for injections, jailbreaks, system-prompt exposure, and sensitive terms. The project is open source, supports self-hosting, and offers cloud plans including a 1,000-detection monthly allowance.
Review
In a line
Prompt Inspector is an open-source screening layer for AI applications and agents. It screens incoming prompts before they reach a language model.
What it does
The service checks for injection attacks, jailbreak attempts, system-prompt leaks, and sensitive content. Results return a safety decision, score, threat categories, request ID, and processing time.
Its detection funnel starts with cached results and word or pattern checks, then uses semantic similarity for paraphrased attacks. Unclear inputs can receive an extra AI verdict and score arbitration. Tenants can set their own sensitive-word lists and regular-expression patterns.
How it connects
Clients can send JSON to a REST endpoint with an X-App-Key header. Official Python and Node.js packages are provided, and the MCP server brings detection to Cursor, VS Code Copilot, Claude Desktop, and Dify.
Agent skill scripts support OpenClaw and Claude Code. API keys can be passed directly or through the PMTINSP_API_KEY environment variable. MCP also accepts X-App-Key or Bearer credentials.
The core project is released under AGPL-3.0. Documentation describes non-commercial self-hosting with Python 3.11 or newer, PostgreSQL 15 or newer with pgvector, Redis 7 or newer, and Node.js 18 or newer.
Good fit
It suits teams adding a security check to an AI app, agent, or API workflow. The maker says every app receives separate credentials, sensitive-word settings, rate limits, and detection records, keeping tenant data apart.
The staged checks are useful when obvious threats should be handled before slower semantic or AI review. Enterprise customers can add private or on-premise deployment, custom integrations, analytics, audit logs, team roles, SSO/SAML, onboarding, and model fine-tuning.
Watch for
Inputs are limited to 5,000 characters. Documentation notes that the optional AI review is slower and more expensive, so it is used for gray-zone cases.
Cloud billing is described as pay-as-you-grow, with no price stated on the opened homepage. Published Free allowances differ: the homepage says 1,000 detections a month, while an API article lists 10,000 requests a month and 10 requests per second.
Self-hosted users receive community support. Check the deployment effort, license terms, and which allowance applies to your account before choosing a plan.
Pricing
The Free tier costs nothing and lists 1,000 detections per month; no credit-card requirement is stated.
Starter has request-based pricing and lists 100,000 requests a month, 50 requests per second, and under 50 milliseconds at the 95th percentile.
Enterprise pricing is available on request and covers private or on-premise deployment plus dedicated support.
Verdict
Choose Prompt Inspector if you need prompt-injection checks, custom sensitive-word rules, and several ways to connect an AI application. Its open-source edition and self-hosting option give non-commercial users control over deployment, while paid tiers cover larger or private setups.
Look elsewhere if a fixed public price, a simpler hosted setup, or inputs longer than 5,000 characters are essential. Confirm the monthly limit and support level in writing, especially because the published Free limits differ.
Our take
Prompt Inspector is a sensible match for builders who want one screening layer across APIs, SDKs, MCP clients, and agent tools. Its open-source code and non-commercial self-hosting option may suit organizations seeking deployment control. The main buying checks are the 5,000-character input cap, the infrastructure needed to run it yourself, and unclear cloud pricing. Check whether your account uses the homepage’s 1,000 detections or the API article’s 10,000-request allowance before depending on the Free plan.
- AGPL-3.0 edition with non-commercial self-hosting
- Screens for injections, jailbreaks, leaks, and sensitive material
- REST API plus Python and Node.js SDKs, MCP server, and agent skills
- Free tier lists 1,000 detections per month
- Separate app credentials, rules, limits, and detection records
- Input length is capped at 5,000 characters
- AI review is slower and more expensive
- Published Free limits differ between pages
- Self-hosting needs Python, PostgreSQL, Redis, and Node.js
- Pick it ifTeams securing prompts in AI apps and agents; Organizations wanting open-source or self-hosted deployment; Workflows using REST, SDKs, MCP, or agent skills
- Skip it ifYou need a fixed public cloud price; Your inputs regularly exceed 5,000 characters; You want a managed service without infrastructure work
Pricing
The Free tier lists 1,000 detections a month with no stated credit-card requirement. Pro uses pay-as-you-grow cloud billing, with its price and allowance unstated. Starter and Enterprise use request-based pricing; Starter lists 100,000 requests a month, while Enterprise adds private or on-premise deployment and dedicated support.
- Free1,000 detections/month · no credit card statedFree
- EnterprisePrivate/on-premises deployment · dedicated supportOn request
- Propay-as-you-grow · Cloud service; exact allowance and price not statedOn request
- Starter100,000 requests/month · 50 req/s · < 50 ms P95On request
As appsecsanta.com lists them, checked 15 Oct 2026.
Questions
What threats does Prompt Inspector detect?
Its engine targets prompt injections, jailbreaks, system-prompt exposure, sensitive-content leaks, plus tenant-defined words and regular-expression patterns.
Does Prompt Inspector have a free plan?
Yes. The homepage gives the Free tier 1,000 detections per month and does not state a credit-card requirement. An API article gives a different Free limit of 10,000 requests a month and 10 requests per second.
Can I run Prompt Inspector myself?
Yes, non-commercial deployments are permitted under AGPL-3.0. The documented stack uses Python 3.11 or newer, PostgreSQL with pgvector, Redis, and Node.js 18 or newer.
Which tools can connect through MCP?
The identified MCP clients are Cursor, VS Code Copilot, Claude Desktop, and Dify. The server exposes a detection tool over Server-Sent Events.
What are the cloud plan prices?
Pro, Starter, and Enterprise show pricing on request. Pro uses pay-as-you-grow billing; Starter lists 100,000 requests a month; Enterprise includes private or on-premise deployment and dedicated support.
How long can an input be?
Repository settings impose a 5,000-character maximum input length.
Near it
All alternatives →- NN
NVIDIA NeMo Guardrails#2 on this board - CCyberSecEval#3 on this board
- P
PromptGuard#4 on this board - SRSanctum Runtime#5 on this board
- OGOpenlayer Guardrails#6 on this board
- G
GuardionAI#7 on this board - VVotalAI#8 on this board
- GA
Guardrails AI#9 on this board - IInferwall#10 on this board
- GGLACIS#11 on this board