Traces, evals, prompt management and metrics to debug and improve your LLM application.
Learn more here
Below is an outline of the steps involved in a simple Math Agent. Key elements illustrated include:
- A visual breakdown of each step—e.g., when the agent invokes a tool and when control returns to the agent
- Inputs and outputs at each stage of the process:
- User to Agent: The user asks a natural language question — e.g.,
What is 1 + 1? - Agent to Tool: The agent decides to call a calculator tool with structured arguments — e.g.,
args: { a: 1, b: 1 }. - Tool to Agent: The tool executes the operation and returns the result — e.g.,
2. - Agent to User: The agent responds with the final answer in natural language — e.g.,
1 + 1 = 2.
- User to Agent: The user asks a natural language question — e.g.,
- The full chat history throughout the interaction
- Latency and cost associated with each node
This section shows:
- User to Agent: The user asks a natural language question — e.g.,
What is 1 + 1? - Agent to Tool: The agent decides to call a calculator tool with structured arguments — e.g.,
args: { a: 1, b: 1 }. - Full Chat History Throughout the Interaction: You can inspect earlier user-agent messages. For instance:
User: reply only no
Agent: No.
In this example, the agent responded directly without calling any tools.
This section shows:
- Tool to Agent: The tool executes the operation and returns the result — e.g.,
2.
This section shows:
- Agent to User: The agent responds with the final answer in natural language — e.g.,
1 + 1 = 2.