Cossistant is a vertical application of the AI agent pattern specifically for the customer support domain. It occupies the 'action' and 'integration' layers of the agent stack, moving beyond simple RAG (Retrieval-Augmented Generation) to allow agents to interact with third-party tools like Stripe and Linear.
For the agent ecosystem, Cossistant is an example of the human-in-the-loop (HITL) architecture. It treats human intervention as a primary data source for agent training, rather than a failure state. By providing a structured way to identify users and pass application context through React props, it solves one of the hardest problems in agent deployment: giving the agent enough specific, verified context to be useful without compromising security or user privacy.
Cossistant is a departure from the traditional support widget. Most customer service platforms operate as third-party scripts that developers drop into their HTML, resulting in a black-box experience where the UI and data flow are difficult to control. Cossistant is a framework built specifically for React and Next.js applications, distributed as an NPM package. This approach allows the support interface to live inside the application's existing codebase, sharing the same state and authentication context as the rest of the product.
The technical implementation relies on a specific set of components, such as <IdentifySupportVisitor />, which allows the agent to ingest rich metadata about a user, including their subscription tier, billing history, or recent activity. This means the agent isn't just a chatbot guessing based on documentation; it is an integrated part of the application logic that understands exactly who it is talking to.
The core of the product is a feedback loop designed to reduce support volume without sacrificing accuracy. When a customer asks a question that the AI cannot answer confidently, the system is programmed to escalate to a human immediately rather than making up information. Once a human team member provides a response, Cossistant indexes that answer. The next time a similar query arrives, the agent uses the historical human response to resolve the issue.
This "learn once, answer forever" model shifts the burden from constant manual ticket resolution to a supervised training exercise. It effectively turns the support inbox into a growing knowledge base. The dashboard gives team members visibility into which questions are being handled by AI and which require intervention, providing a metrics-driven view of support efficiency.
Beyond simple text responses, Cossistant supports "skills" or programmable actions. This is where the framework crosses the line from a chatbot into an agent. By connecting to external APIs, the agent can perform tasks such as checking a user's subscription status in Stripe, logging a bug in Linear, or booking a meeting via Cal.com.
Because it is code-first, developers can wire up custom actions specific to their own internal APIs. If a user needs to rotate an API key or replay a failed webhook, the agent can be granted the specific "skill" to perform those actions after verifying the user's identity through the application's native auth system.
Cossistant is built by Marc Louvion and Nico Jeannen, two prominent figures in the indie developer community known for rapid product iteration. Based in California City and operating with a small, remote-first team, the company targets SaaS founders who want to scale support without hiring a massive team.
By keeping the framework open-source and developer-focused, they are competing with the complexity and high cost of enterprise-level support suites. Their value proposition is technical: if you can write ten lines of React code, you can have a support agent that understands your product as well as your lead developer does.
AI agent customer support for your SaaS in under 10 lines of code.
Cossistant is hiring