Investment research is tedious and non sequential, which makes it costly and non automatable. This engine aims to allow for a powerful and cost efficient way of doing financial related research. It combines AI and external data services through an agent based operating model.
An orchestrator agent articulates a research path. This path is built by spawned sub-agents with a dedicated goal. These agents can query support agents to access specific data sources. They should also be able to spawn their own sub-agents.
- The
main.gofile is the main file. It allows to launch the web server. Adding the -cli flag followed by the name of company will launch the workflow from the terminal. - The orchestrator.go file handles the triggering of the workflow by spawning the main agent which uses the
OrchestratorPrompt.mdand theSystemPrompt.md - The
/toolsfolder is the external tools folder (support agents and data sources). It contains the workflow allowing agents to access external data. - The logic to satisfy a specific need through a tool is the following
PickTool.go(Agent) >PrepareRequest.go(Agent) >ExecuteRequest(Hard coded) >EvaluateResponse.go(Agent) - The
/tools/tools.xmlfile allows to list all the tools available as well as their description and the path to their documentation. It's used by thePickTool.goagent. - Each tool has its own subfolder with two files :
toolName.md, which is a generic documentation on the tool and its access, and often atoolName.jsonswagger. - In the future, a failure to satisfy a data need should lead to an automatic ticket creation by the agent.
- Clone the repo locally
git clone https://github.com/Mathiasme/ClaudeShorts - Create an .env file at the root of the folder. Add an ANTHROPIC_API_KEY. You'll also need to add keys required by each support tool used by the engine.
go run main.go
There is an infinite reasons to why a stock might go up, many of them are subjective and unpredictable. On the opposite reasons on which a stock might go up are fewer but offer a stronger signal.