Pangea AI Guard

Free plan APIExtensionSelf-hosted

What Pangea AI Guard is

Pangea AI Guard protects data and interactions in applications that use large language models. It detects prompt injection, malicious entities, personal and confidential information, secrets, language, code, competitors, custom entities, and restricted topics. Early access features cover images and prompt hardening. Teams can combine detectors into recipes and choose whether a detection is reported, blocked, redacted, encrypted, or defanged. Recipes can be created for prompts, RAG ingestion, LLM responses, and agent tool inputs or outputs. Detection uses heuristics, machine-learning classifiers, and fine-tuned LLMs. Applications can connect through APIs and SDKs for Python, Node.js, Go, Java, and C#, or use gateway integrations for Portkey, Kong, and LiteLLM. Pangea offers SaaS, Edge, and Private Cloud deployment; Edge and Private Cloud process data within the customer’s cloud boundary. Enabling AI Guard activates an Activity Log for service calls, inputs, outputs, detections, and contextual information. Webhooks can notify users about selected detections. The current version processes up to 20 KiB of text per request. A free Pangea account can enable AI Guard and its APIs.

Who it suits

AI Guard suits teams building applications that interact directly with LLMs, exchange data through agents, or use RAG ingestion and retrieval. It is also relevant to teams that need configurable detection actions, audit logging, or deployment within their cloud boundary.

What it does well

Detects prompt injection, PII, secrets, and other configurable categories.
Recipes can report, block, redact, encrypt, or defang content.
Supports APIs, five SDK languages, and three gateway integrations.
Activity Log records service calls, inputs, outputs, detections, and context.
Edge and Private Cloud process data within the customer’s cloud boundary.

What to know first

Requests are limited to 20 KiB of text.
Kong integrations do not currently support streaming responses.

Limits

The free Pangea account can enable AI Guard and its APIs. Each request supports up to 20 KiB of text; listed Kong plugins do not currently support streaming responses.

Getting started

Open the Pangea AI Guard documentation website.
Create a free Pangea account.
Choose an API or SDK integration, or use a documented gateway integration.
Select SaaS, Edge, or Private Cloud deployment.
Configure detectors and recipes for the relevant application stages.

Verdict

Choose Pangea AI Guard if an application needs configurable LLM threat and sensitive-data detection, with options for audit logging and deployment within a customer cloud boundary. Its request limit is 20 KiB of text, and streaming responses are not currently supported for the listed Kong plugins.

Questions

Is there a free plan?

Yes. A free Pangea account is sufficient to enable AI Guard and start using its APIs.

Which platforms and SDK languages are supported?

The listed platforms are API, extension, and self-hosted. SDKs are documented for Python, Node.js, Go, Java, and C#.

What can AI Guard detect?

Detectors cover prompt injection, malicious entities, PII and confidential data, secrets, language, code, competitors, custom entities, and restricted topics. Image and prompt-hardening features are early access.

Which integrations are available?

Applications can connect through APIs and SDKs, or through gateway integrations for Portkey, Kong, and LiteLLM.

What does the Activity Log record?

When AI Guard is enabled, the Activity Log records service-call details, inputs, outputs, detections, and contextual information.

Who makes AI Guard?

Pangea makes AI Guard. Its listed headquarters address is 636 Ramona St, Palo Alto, CA 94301.

Facts from Pangea AI Guard's own pages

Alerts: Users can configure webhooks and link them to specific detections to receive activity notifications and take action. · pangea.cloud, 2 Oct 2026
Audit logging: When AI Guard is enabled, its Activity Log is automatically activated and records service-call details including input, output, detections, and contextual information. · pangea.cloud, 2 Oct 2026
Audit trail: AI Guard automatically activates an Activity Log that records service calls, inputs, outputs, detections, and contextual information in the project audit trail. · pangea.cloud, 3 Oct 2026
Company: Pangea’s company page lists its address as 636 Ramona St, Palo Alto, CA 94301. · pangea.cloud, 2 Oct 2026
Configurable actions: Recipes combine configurable detectors that can report, block, redact, encrypt, or defang detected content. · pangea.cloud, 2 Oct 2026
Deployment: Pangea lists SaaS, Edge, and Private Cloud deployment models; Private Cloud runs data and control planes in the customer's infrastructure and is marked early access. · pangea.cloud, 3 Oct 2026
Detection: Its detectors cover prompt injection, malicious entities, PII and confidential data, secrets, language, code, competitors, custom entities, restricted topics, and early access image and prompt-hardening features. · pangea.cloud, 2 Oct 2026
Detection and sanitization: It detects prompt injection and other threats, filters content, and can redact or encrypt sensitive information in submitted text. · pangea.cloud, 3 Oct 2026
Detection methods: Threat and sensitive-data detection uses heuristics, machine-learning classifiers, and fine-tuned LLMs. · pangea.cloud, 2 Oct 2026
Detector coverage: Available detectors cover malicious prompts and entities, PII and confidential data, secrets, language, code, competitors, custom entities, restricted topics, and early-access image and prompt-hardening features. · pangea.cloud, 3 Oct 2026
Free account: Pangea’s getting-started guide says a free Pangea account is sufficient to enable AI Guard and start using its APIs. · pangea.cloud, 2 Oct 2026
Gateway integrations: Documented AI and API gateway integrations include Portkey, Kong, and LiteLLM. · pangea.cloud, 2 Oct 2026
Headquarters: 636 Ramona St, Palo Alto, CA 94301 · pangea.cloud, 28 Sept 2026
Input limit: The current version supports processing up to 20 KiB of text. · pangea.cloud, 2 Oct 2026
Integration: Applications can connect through APIs or SDKs, and AI Guard documents SDKs for Python, JavaScript (Node.js), Go, Java, and C# through its integration options. · pangea.cloud, 2 Oct 2026
Integrations: AI Guard can be integrated through APIs and SDKs or through gateway extensions and plugins for Portkey, Kong, and LiteLLM. · pangea.cloud, 3 Oct 2026
Intended users: Pangea describes AI Guard as protection for AI-powered applications, including direct LLM interactions, agent data exchanges, and RAG data ingestion and retrieval. · pangea.cloud, 3 Oct 2026
Pricing model: Pangea’s billing page lists AI Guard pricing details by data scanned, but the opened page showed only “Loading billing information...” and no price amount. · pangea.cloud, 2 Oct 2026
Purpose: AI Guard protects data and interactions with LLMs in AI-powered applications by safeguarding sensitive information, blocking malicious content, and recording application events. · pangea.cloud, 3 Oct 2026
Recipes: Users can customize recipes made of detectors or create an unlimited number of recipes for stages such as prompts, RAG ingestion, LLM responses, and agent tool inputs and outputs. · pangea.cloud, 3 Oct 2026
Request limit: The current version supports processing up to 20 KiB of text per request. · pangea.cloud, 3 Oct 2026
Security certifications: The Pangea site displays SOC 2 Type 2, ISO/IEC 27001, and ISO/IEC 27701 compliance badges. · pangea.cloud, 3 Oct 2026
Support: Pangea lists [email protected] for general inquiries and [email protected] for technical support. · pangea.cloud, 3 Oct 2026
Supported gateway providers: The Kong plugins list Anthropic Claude, Azure OpenAI, Cohere, Google Gemini, Kong AI Gateway, and OpenAI as supported providers, and state that streaming responses are not currently supported. · pangea.cloud, 3 Oct 2026

Where Pangea AI Guard stands on Freedom251

AI Agent Security Software#6 of 31
AI Guardrail Software#12 of 23