Frigade is adding a new layer to enterprise AI agents: direct knowledge of how a software product actually works. The company has launched the Assist API, a framework-agnostic tool that lets existing AI agents answer product questions and guide users through live software interfaces without requiring companies to deploy a separate assistant.
Frigade’s new Assist API targets a growing problem for companies building AI-powered customer experiences: an agent can be excellent at generating language while still knowing surprisingly little about the software it is supposed to explain.
The San Francisco-based company says its new API gives an organization’s existing AI agent access to product expertise gathered from the software itself. Instead of relying exclusively on a help center, documentation or static retrieval-augmented generation (RAG) pipeline, Frigade’s system interacts with a customer’s product through a browser and learns its workflows, features and relationships.
The distinction matters as enterprise AI moves from chatbots toward agents capable of taking actions and guiding people through multi-step workflows.
Frigade says its browser-based agent signs into a customer’s application using the same permissions as a normal user, works through real product flows and builds a model of how the application operates. The company repeats that process as the product changes, allowing its knowledge to track newer releases rather than remaining tied to documentation written months earlier.
That approach addresses a familiar weakness in AI-powered support. Documentation can become outdated as interfaces, navigation and features change. A conventional AI assistant may retrieve an accurate article that nevertheless describes an older version of the product, leaving the customer to translate the instructions into the interface in front of them.
The Assist API instead lets the company’s own agent call Frigade when it needs product-specific knowledge. The response can be a conventional answer or an interactive walkthrough that highlights steps directly inside the user’s browser.
Frigade positions this as a way to combine onboarding and support without creating another conversational interface. A software vendor’s existing agent remains the user-facing assistant, while Frigade supplies product knowledge behind the scenes. The company says the API can work with any agent capable of calling a tool, including applications built with the Vercel AI SDK.
That architecture is increasingly relevant to enterprise AI development. Companies are building agents around foundation models from vendors such as Microsoft, Google and Amazon, but the general-purpose intelligence of those models does not automatically provide knowledge of a company’s proprietary application or the exact workflow a customer is using.
The broader market is moving in this direction. Gartner expects worldwide end-user spending on generative AI models to reach $14.2 billion in 2025 and predicts that more than half of enterprise GenAI models will be domain-specific by 2027, compared with 1% in 2024. The trend suggests that enterprises increasingly want AI systems optimized around specific business contexts rather than generic model capabilities.
McKinsey’s 2025 State of AI research points to another piece of the puzzle: 62% of respondents said their organizations were at least experimenting with AI agents, while nearly two-thirds had not yet begun scaling AI across the enterprise. For vendors building those agents, connecting models to reliable enterprise context is becoming as important as selecting the underlying LLM.
Frigade’s approach also puts product permissions into the architecture. The company says guidance operates using each user’s existing permissions, meaning the agent should only access information that the user could already see. Enterprise deployments can also be self-hosted using a customer’s own LLM keys. Frigade says it is SOC 2 Type II certified and GDPR compliant, with EU data residency available.
The company is extending technology it already uses in Frigade Assistant, its in-product AI system, and integrating it with Frigade Engage, which provides product tours, checklists and announcements. Support teams can review conversations handled by the agent, correct responses and modify its knowledge without requiring engineering changes, according to the company.
Retell AI, a voice AI infrastructure provider, is among the companies building on the Assist API, according to Frigade.
The competitive significance is less about another AI chatbot entering the market and more about where enterprise AI agents get their operational context. Salesforce, Adobe and other enterprise software companies are embedding AI directly into business applications, while infrastructure vendors are making it easier for developers to build agents around existing LLMs. Frigade is betting that a complementary product-knowledge layer can sit underneath those experiences.
That creates a potentially useful separation of responsibilities: the enterprise owns the agent and the customer relationship, the LLM provides general reasoning and language capabilities, while a specialized system supplies current knowledge of the application’s interface and workflows.
The remaining challenge will be proving that automatically learned product knowledge remains accurate enough for production support, particularly as applications become more complex and agents move from explaining workflows to executing them. For enterprise AI, that distinction could become increasingly important: knowing what a product can do is one problem; reliably knowing what a specific user is allowed to do is another.
Market Landscape
The Assist API enters a market where enterprise AI is shifting from standalone chat interfaces toward AI agents, embedded copilots and workflow automation. Microsoft, Google, Amazon, Salesforce and other major platforms are integrating agents into enterprise software, while developers increasingly assemble their own agent stacks around LLMs and AI development frameworks.
Frigade occupies a narrower layer: product-aware AI infrastructure. Its pitch is that an agent should understand the current application rather than depend solely on static documentation or generic model knowledge.
The timing aligns with two major trends. Gartner forecasts $14.2 billion in enterprise end-user spending on GenAI models in 2025 and expects domain-specific models to become much more common. Meanwhile, McKinsey reports that AI adoption is widespread, but enterprise-scale deployment remains relatively immature.
This leaves room for infrastructure that improves the connection between LLMs and proprietary enterprise workflows.
Top Insights
- Frigade’s Assist API gives existing AI agents access to live product knowledge instead of forcing companies to deploy another customer-facing assistant.
- The platform learns workflows through browser-based product interaction, helping reduce reliance on documentation that can become outdated.
- Framework-agnostic tool calling allows the technology to work with existing agent architectures, including applications built with the Vercel AI SDK.
- Enterprise customers can self-host Frigade with their own LLM keys, addressing data-control and compliance requirements.
- The launch reflects a broader shift toward domain-specific AI agents that understand business applications, permissions and operational workflows.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
