Tool registry, function schemas, tool calls, and tool results — the vocabulary of how agents interact with the world.
Key vocabulary
Tool registry — the list of tools available to an agent, with their schemas.
Function schema — the description of a tool: name, description, parameters, and types.
Tool call — when the agent invokes a tool with arguments.
Tool result — the output returned to the agent after a tool executes.
Tool chaining — using the output of one tool as input for the next.
0 / 5 completed
1 / 5
What is a tool registry in an agentic system?
Tool registry = the catalogue of tools the agent can access. Each entry includes the tool’s name, description, and parameter schema. The agent reads this registry to decide which tool fits the current sub-task.
2 / 5
A function schema in a tool definition serves to:
Function schema = the structured description of a tool the LLM reads. It includes: name, description (plain language), and parameters (types, descriptions, required/optional).
3 / 5
In a tool use cycle, after the agent sends a tool call, the next step is:
The cycle: agent calls tool → tool executes → tool result returned → agent observes → agent reasons next step. The tool result is a first-class message the agent reasons from.
4 / 5
When an agent chains multiple tool calls — search, then read a URL, then summarise — this is called:
Tool chaining = using the output of one tool as input for the next. Complex tasks require sequences: search → filter → read → summarise.
5 / 5
A tool call with invalid arguments typically results in:
Tool errors are normal. A well-designed agent reads the error, corrects the arguments, and retries. Good function schemas with clear descriptions reduce invalid calls.