The Problem
Finance Guru addresses the fragmentation of personal financial analysis—prices in one tab, spreadsheets in another, risk models in a third. The project's core claim is that a coordinated team of AI agents can replace that workflow with a single command that routes a question to the appropriate specialist and returns a consolidated answer.
What This Does
Finance Guru is a two-layer system. The fin-guru/ directory defines the agent orchestration layer: YAML configs in framework/ (routing, directives, knowledge base) plus task definitions and agent role prompts in agents/ and tasks/. These are Claude Code command definitions, not standalone software—activating /finance-orchestrator loads a multi-agent workflow.
The apps/plaid-dashboard/ directory is the actual application code: a Next.js dashboard (dashboard/) for account/transaction views, a TypeScript engine (engine/) with Plaid provider routes (routes/accounts.ts, routes/sync.ts, routes/transactions.ts), and a Drizzle ORM database layer (db/src/schema.ts). The dashboard has 13 test files and uses Docker Compose for local infrastructure.
How To Use It
Setup and usage depend on which layer you're targeting:
Agent layer (Claude Code)
Requires Claude Code CLI; invoke the orchestrator from the repo root /finance-orchestrator
Dashboard application
cd apps/plaid-dashboard cp .env.example .env # requires PLAIDCLIENTID, PLAIDSECRET, DATABASEURL docker compose up -d # database npm install npm run dev # Next.js dashboard
Configuration lives in .env.example for the dashboard and fin-guru/config.yaml for agent behavior. The agent layer's docs (docs/index.md, docs/api.md) describe workflows but do not document a standalone CLI entry point—the Python analysis tools referenced in the README (src/utils/momentum_cli.py) are not present in the file listing, so those commands cannot be run as documented.
Real-World Use
A user asks the orchestrator whether to add a position. The agent layer routes to the Quant Analyst and Compliance Officer, who evaluate the request against the portfolio context loaded from fin-guru/tasks/load-portfolio-context.md. For the dashboard, a user connects a bank via components/plaid-link-button.tsx, which exchanges a public token through actions/exchange-public-token.ts; the engine's sync-service.ts then pulls transactions into the database for display in components/transactions-table.tsx.
Code Health & Issues
High – Missing CI/CD: No .github/ or CI config exists. With 13 tests present, there is no automated gate to run them on commit or PR. Medium – No license: The README marks "Private" but no LICENSE file exists. This blocks external contribution and reuse. Medium – Documentation/reality mismatch: README references src/analysis/ and src/utils/ Python tools that are absent from the file tree. Either the docs are stale or the analysis layer is missing. Medium – Monorepo inconsistency: The .cursor/rules/use-bun-instead-of-node-vite-npm-pnpm.mdc mandates Bun, but the dashboard uses npm (package.json, package-lock.json). The root has both bun.lock and npm artifacts. Low – Duplicate schema definitions: db/src/schema.ts and engine/src/db/schema.ts may drift; the engine imports its own copy rather than the shared package.
The Bottom Line
Finance Guru is an ambitious agent-orchestration framework with a functional Plaid dashboard as its application layer. The agent system is well-structured for Claude Code users who want a repeatable multi-agent financial workflow. The dashboard is a credible starter, but the missing analysis tools, absent CI, and no license make it unsuitable for production deployment without significant hardening. Best suited for a technical individual exploring AI-driven financial workflows, not a team building a regulated product.