llm4s-docs
API
Generated with
llm4s-docs/org.llm4s/org.llm4s.agent/AgentNode

AgentNode

org.llm4s.agent.AgentNode
enum AgentNode

A node of an agent's loop that a static breakpoint can hold: see AgentBuilder.withInterruptBefore and AgentBuilder.withInterruptAfter. Each task the breakpoint holds is an interrupt of its own, reported in AgentStatus.Suspended.breakpoints and continued with AgentResult.proceed.

A step continued after a review runs at another node: a model call after a wrapModelCall question at <id>/asked/<middleware>/wrapModelCall, a tool call after an approval or a question at <id>/approval, <id>/ask/<tool> or <id>/asked/<middleware>/wrapToolCall, and an answer after an afterAgent question at <id>/asked/<middleware>/afterAgent. A breakpoint after a node holds these too, so nothing the step does escapes it; a breakpoint before one does not, since the reviewer has just seen the step.

Attributes

Graph
Supertypes
trait Enum
trait Serializable
trait Product
trait Equals
class Object
trait Matchable
class Any
Show all

Members list

Type members

Enum entries

case Finish extends AgentNode

The final answer (<id>/finish), before its afterAgent hooks - guardrails among them - run.

The final answer (<id>/finish), before its afterAgent hooks - guardrails among them - run.

Attributes

case Model extends AgentNode

Each model call (<id>/model). Held before, the model is not called yet; held after, the model's message is stored, but its tool calls - or its final answer's guarding - wait.

Each model call (<id>/model). Held before, the model is not called yet; held after, the model's message is stored, but its tool calls - or its final answer's guarding - wait.

Attributes

case Tool extends AgentNode

Each tool call (<id>/call-tool), one interrupt per call, so the calls of one message are held and continued one at a time. Held before, the call has not run; held after, its result is recorded, but the next model call waits for it - also a result recorded once an approval or a question about the call was answered.

Each tool call (<id>/call-tool), one interrupt per call, so the calls of one message are held and continued one at a time. Held before, the call has not run; held after, its result is recorded, but the next model call waits for it - also a result recorded once an approval or a question about the call was answered.

Attributes

Generated with