Omnigent's multi-agent workflow lets you delegate financial summaries to Claude—here's how

Building an AI agent that actually does something useful—like fetching a live exchange rate and handing you a clean summary—usually means wrestling with multiple frameworks, API keys, and deployment headaches. A new tutorial from Omnigent, the open-source agent orchestration system, shows a cleaner path. It walks through constructing a two-agent financial research workflow entirely inside a Colab notebook, using nothing more than YAML config files, Python functions, and the Claude Agent SDK as the execution engine.

The setup is centered on a lead research agent. When asked about currency movements, it calls a live API to grab the current USD-to-EUR rate, drafts a summary, and then automatically delegates that draft to a second agent—a text auditor. The auditor runs a word count, flags jargon, and suggests clarity improvements before the lead agent delivers the final version. Both agents are defined in a single YAML file, with their tools mapped to plain Python functions like get_exchange_rate and word_count.

What makes the tutorial noteworthy isn’t the exchange-rate lookup itself—that’s a five-line function hitting a free API. It’s the governance layer. The workflow enforces non-interactive policies that limit tool calls and cap session costs, and the Anthropic API key is handled through environment variables rather than hardcoded strings. The whole thing runs in an isolated virtual environment built with uv, avoiding the dependency hell that often plagues Colab setups. “We also manage the Anthropic API key securely through environment variables, apply non-interactive policies that limit tool calls and control session costs, and run the workflow directly from Colab without requiring Node.js, tmux, or an interactive terminal,” the tutorial notes.

This pattern—defining agents, their tools, and their delegation rules in a single configurable spec—hints at where agent development is heading. Less scaffolding code, more declaring what you want the system to do. The workflow produced here is simple, but the architecture scales: swap the functions, change the prompts, and you’ve got a different multi-agent system without touching the orchestration layer.