Frigade Assist API

Product Introduction

1. Definition: The Frigade Assist API is a specialized API and SDK designed for AI agent tooling. It functions as a single tool call that can be integrated into an existing AI agent (like one built with the Vercel AI SDK) to provide that agent with real-time, contextual expertise about a specific software product.
2. Core Value Proposition: It exists to solve the critical gap where AI agents lack contextual awareness of a live application interface. Instead of providing generic text-based answers from potentially outdated documentation, the Frigade Assist API enables agents to "see" the user's screen and deliver actionable, step-by-step guidance rendered directly on the product's UI, or provide answers grounded in the product's current state.

Main Features

1. Contextual In-App Guidance: The API allows an agent to trigger interactive, step-by-step guides that are overlaid directly onto the user's application interface. This works by the agent calling the Frigade tool with a user query; Frigade then identifies the relevant workflow and draws highlights, tooltips, and navigation cues on the actual UI elements the user needs to interact with.
2. Autonomous Product Learning & Relearning: Frigade learns a product by deploying browser-based agents that autonomously navigate and interact with the application as a real user would. It maps out workflows, UI states, and feature connections. Crucially, this process repeats with every software release, ensuring the internal product model is always synchronized with the latest version, eliminating documentation drift.
3. Grounded Answer Generation: When live guidance isn't possible or necessary, the API provides plain-text answers that are generated from its continuously updated model of the product. These answers are based on how the product actually behaves in its current version, not on static help articles. This feature includes a feedback loop where support teams can rate answers and provide corrections without writing code.
4. Framework-Agnostic Tool Integration: The core offering is a simple JavaScript/TypeScript SDK (@frigade/ai) that exposes a single frigade.assist({ query }) function. This can be wrapped as a standard tool definition and added to virtually any AI agent framework, as it relies only on the fundamental ability to make a function call. It is explicitly compatible with the Vercel AI SDK but is not limited to it.

Problems Solved

1. Pain Point: Agent Hallucination and Outdated Information. Traditional AI agents trained on documentation become inaccurate immediately after a product update, leading to wrong answers and user frustration.
2. Pain Point: Ineffective Text-Only Support. Answering "how do I do this?" with a wall of text or a link to a help doc forces users to context-switch and interpret instructions, leading to a poor support experience and increased support tickets.
3. Target Audience: Product & Engineering Teams building AI-powered customer support or onboarding assistants. Developer Advocates and Technical Product Managers responsible for user education and feature adoption. Support & CX Leaders seeking to scale their team's impact with automation while maintaining control over answer quality.
4. Use Cases: In-Agent Product Support: An AI assistant in a SaaS dashboard can now guide a user through configuring a webhook in real-time. Proactive Feature Adoption: An agent can identify a user's need and use Frigade to proactively launch a guided tour of a relevant, underused feature. Accurate Q&A: An agent can provide definitive, version-accurate answers to complex billing or plan feature questions without human intervention.

Unique Advantages

1. Differentiation: Unlike static documentation platforms (e.g., Intercom Articles, Zendesk Guide) or manual tour builders (e.g., Userpilot, Appcues), Frigade's core innovation is its autonomous, behavioral learning engine. Competitors require manual configuration and constant updating by humans. Frigade's system learns and updates itself by interacting with the live application.
2. Key Innovation: The "Relearns on Every Release" capability. The product treats the application itself as the single source of truth. By automating the discovery and mapping process post-deployment, it ensures the knowledge base is perpetually current. This is a fundamental shift from a document-centric to a product-behavior-centric model for AI training.

Frequently Asked Questions (FAQ)

1. How does the Frigade Assist API actually learn my product's workflows? Frigade uses automated browser agents that are granted access to your application (similar to a test user). These agents systematically navigate through your product's UI, clicking buttons, filling forms, and traversing workflows to build a dynamic map of how features connect and operate in the current version.
2. Is the Frigade Assist API secure for use in production applications? Yes. Frigade is SOC 2 Type II certified and GDPR compliant. It uses TLS 1.2+ encryption in transit and AES-256 at rest. The guidance runs with the end-user's own permissions and session, meaning the agent only accesses what that user can already see. For maximum control, Frigade offers a self-hostable option.
3. Can I use the Frigade Assist API if I don't have an AI agent yet? Yes. Frigade offers a standalone, no-code Frigade Assistant product that provides a complete, pre-built in-product AI assistant interface. The Assist API is specifically for teams who have already built a custom agent and want to augment it with Frigade's product intelligence capabilities.
4. What happens if Frigade doesn't know how to answer a user's question? The API is designed to fail gracefully. If it cannot generate a grounded answer or guide, it will explicitly state its limitations to the calling agent. This allows your agent to maintain control of the conversation and either hand off to a human or try an alternative approach without stalling.
5. How is pricing structured for the Frigade Assist API? Frigade operates on a subscription model starting at approximately $1,000 per month, with usage-based scaling. This typically includes a set number of monthly assist calls (tool invocations), with enterprise plans available for higher volumes, advanced security needs, or self-hosting requirements.