agutinbaigo28/financial-agent-api

AgentStack is a platform for building and running multiple AI agents that work together, built on a framework called Mastra. It combines many agents, tools, and workflows into one system so they can coordinate on tasks instead of acting as one single chatbot. The project focuses on financial data and analysis, along with the infrastructure needed to run agents in a business setting. The README lists a large number of built in pieces: more than 25 individual agents, 57 tools those agents can call, over 10 multi step workflows, and more than 12 supervisor networks that coordinate groups of agents. There are also over 100 user interface components for building chat style interfaces. Financial intelligence is one of the main features, with tools that pull data from services like Polygon, Finnhub, and AlphaVantage, covering things like stock quotes, company fundamentals, and technical indicators. For working with documents and knowledge, AgentStack includes a retrieval system, often called RAG, that stores information as embeddings in a database called LibSQL, so agents can search and pull in relevant text when answering questions. It also has dataset management features for tracking experiments and evaluating how well the agents perform over time. On the technical side, the project is written mostly in TypeScript and uses Next.js and React for its interface, with Zod for data validation and Vitest for testing, and the README states it has 97 percent test coverage. It also includes security features such as authentication, role based access control, and protections against common web vulnerabilities like path traversal. Observability tooling lets developers trace what each agent, tool, and workflow step is doing, which helps with debugging when several agents are working at once. The project is licensed under MIT, meaning it can be used, modified, and reused freely, including for commercial purposes. It positions itself against other AI agent platforms by emphasizing production readiness: real time tracing, dataset versioning, and coordinated supervisor networks rather than a single simple chatbot. The full README is longer than what was shown.