Agent / LLMCall

Agent.create(options)

Agent.create(options: { provider: LLMProvider; name?: string }): AgentBuilder

Creates an agent builder. Chain methods to configure, then .build() to get a runner.

Builder methods

Method	Description
`.system(prompt)`	Set system prompt
`.tool(tool)`	Add a single tool
`.tools(tools)`	Add multiple tools
`.toolProvider(provider)`	Dynamic tool provider
`.promptProvider(provider)`	Dynamic prompt provider
`.instruction(inst)`	Add conditional instruction
`.decision(initial)`	Set initial Decision Scope
`.pattern(pattern)`	`AgentPattern.Regular` or `AgentPattern.Dynamic`
`.maxIterations(n)`	Max tool loop iterations (default: 10)
`.streaming(bool)`	Enable token streaming
`.memory(config)`	Conversation persistence
`.recorder(rec)`	Attach an AgentRecorder
`.verbose(bool)`	Verbose narrative output
`.build()`	Returns `AgentRunner`

AgentRunner

Method	Returns	Description
`run(message, options?)`	`AgentResult`	Execute the agent
`resume(response)`	`AgentResult`	Resume after pause
`getNarrative()`	`string[]`	Human-readable trace
`getNarrativeEntries()`	`CombinedNarrativeEntry[]`	Structured entries
`getSnapshot()`	`RuntimeSnapshot`	Full execution state
`getSpec()`	`object`	Flowchart specification
`getMessages()`	`Message[]`	Conversation history
`resetConversation()`	`void`	Clear history

AgentResult

interface AgentResult {
  content: string;          // LLM's final response
  messages: Message[];      // Full conversation
  iterations: number;       // Tool loop iterations
  paused?: boolean;         // true if agent paused (ask_human)
  pauseData?: { question: string; toolCallId: string };
}

LLMCall.create(options)

LLMCall.create(options: { provider: LLMProvider }): LLMCallBuilder

Simpler than Agent — single LLM invocation, no tool loop.

Builder methods

Method	Description
`.system(prompt)`	Set system prompt
`.streaming(bool)`	Enable token streaming
`.recorder(rec)`	Attach an AgentRecorder
`.build()`	Returns `LLMCallRunner`

LLMCallRunner

Method	Returns	Description
`run(message, options?)`	`{ content, messages }`	Execute the call
`getNarrative()`	`string[]`	Execution trace
`getSnapshot()`	`unknown`	Full state
`getSpec()`	`unknown`	Flowchart spec