References: How an agent fetches its own data

For education only. Not investment, financial, or trading advice.

Source material

Primary source (open-source, Apache-2.0):
- TradingAgents, an open-source multi-agent LLM framework
  Authors: Yijia Xiao, Edward Sun, Di Luo, Wei Wang
  Organization: TauricResearch
  Repository: https://github.com/TauricResearch/TradingAgents
  Paper: "TradingAgents: Multi-Agents LLM Financial Trading Framework",
         arXiv:2412.20138 (https://arxiv.org/abs/2412.20138)
  License: Apache-2.0
  Pinned snapshot for this course: 7e9e7b8 (the framework as of 2026-05-01)

All code shown in this lesson is quoted from the framework at the pinned
snapshot above, under the Apache-2.0 license, with attribution. Clawdemy's
lesson prose is original. We teach the architecture only; we make no
investment, financial, or trading claims, and we report no performance results.

How to open it (no account or Git needed)

The framework is public and free to read. You do not need a GitHub account, the tool called Git, or any programming knowledge: open the link below in your browser and read the files like ordinary web pages.

• github.com/TauricResearch/TradingAgents (pinned snapshot). This link points to one frozen version of the project, marked by the short code 7e9e7b8.

The files this lesson reads

At the pinned snapshot:

• tradingagents/agents/analysts/market_analyst.py: the analyst’s tools, how they are attached to the model, and the stop condition (no tool call means the prose is the report).
• tradingagents/graph/conditional_logic.py: should_continue_market, the check that decides “run a tool” versus “done.”
• tradingagents/graph/setup.py: the edge from the tool step back to the analyst, which is what makes the loop.
• tradingagents/agents/utils/agent_utils.py: create_msg_delete, the clean-desk message clear between analysts.

Read this next

• TradingAgents repository (TauricResearch). The full source, at the pinned snapshot. The four files above are the ones this lesson quotes.
• TradingAgents paper (arXiv:2412.20138) by Yijia Xiao, Edward Sun, Di Luo, and Wei Wang. The framework’s own description of its design.

Adjacent topics

Where this sits inside this track.

• Why split one AI into many. The previous lesson: the team and why the work is divided into roles. This lesson zoomed in on how one role, an analyst, does its gathering.
• The bull and the bear. The next lesson: the four reports are in, but information is not a decision. Two researchers now argue the case for and against, on purpose and at full strength.
• How tool use turns a model into an agent (foundational agent track). The conceptual predecessor: the general idea of tools and the agent loop, which this lesson shows working in a real system.