Agent

An agent, or one of its conversations. Returned by wardrik.agent.

• Most methods make one request, to the agent.* operation of the same name.
• stream and reply also follow the run that answers a message.
• Methods that can act on a conversation act on conversation.

Extends

• AgentOperations

Properties

agent

readonly agent: string;

The agent's ID.

Inherited from

AgentOperations.agent

---

conversation

readonly conversation: string | undefined;

The conversation the methods act on. undefined for the agent's own history.

Inherited from

AgentOperations.conversation

Methods

stream()

stream(input, options?): AsyncGenerator<WardrikEvent>;

Sends a message and yields the events of the run that answers it, until the run ends.

for await (const event of agent.stream("What's on fire?")) {
  if (event.type === "wardrik.message.delta") process.stdout.write(event.data.delta);
  if (event.type === "wardrik.run.interrupted") {
    for (const interrupt of event.data.interrupts) await agent.approve(interrupt.id);
  }
}

An interrupted run carries on, on the same stream, once each interrupt is approved or denied.

Parameters

Parameter	Type
input	string | Ask
options?	StreamAskOptions

Returns

AsyncGenerator<WardrikEvent>

---

reply()

reply(input, options?): Promise<Reply>;

Sends a message and resolves when the run that answers it ends or waits on a person.

const reply = await agent.reply("What's on fire?");
if (reply.status === "completed") console.log(reply.text);

Parameters

Parameter	Type
input	string | Ask
options?	StreamAskOptions

Returns

Promise<Reply>

---

read()

read(options?): Promise<AgentSnapshot>;

Read what an agent is doing now.

Parameters

Parameter	Type
options?	CallOptions

Returns

Promise<AgentSnapshot>

Inherited from

AgentOperations.read

---

define()

define(input?, options?): Promise<AgentSnapshot>;

Define an agent.

Starts the agent the definition describes, or brings the one running to it from its next turn. A definition put again replaces the last. A key nothing here supports yet answers not_supported, naming it.

Acts on the agent, never a conversation.

Parameters

Parameter	Type
input?	AgentDefinition
options?	CallOptions

Returns

Promise<AgentSnapshot>

Inherited from

AgentOperations.define

---

forget()

forget(options?): Promise<void>;

Forget an agent.

Acts on the agent, never a conversation.

Parameters

Parameter	Type
options?	CallOptions

Returns

Promise<void>

Inherited from

AgentOperations.forget

---

ask()

ask(input, options?): Promise<{
  id: string;
  cursor: string;
}>;

Send an agent a message.

Parameters

Parameter	Type
input	Ask
options?	CallOptions

Returns

Promise<{ id: string; cursor: string; }>

Inherited from

AgentOperations.ask

---

decide()

decide(input, options?): Promise<object[]>;

Approve or reject the calls an agent waits on.

The decisions are taken together and recorded as the caller's. A rejection's reason is what the model is told in place of the call's answer. An approval's note is what it reads beside the answer.

Parameters

Parameter	Type
input	Decisions
options?	CallOptions

Returns

Promise<object[]>

Inherited from

AgentOperations.decide

---

authorization()

authorization(interrupt, options?): Promise<Authorization>;

Read where a person authorizes a call.

Parameters

Parameter	Type
interrupt	string
options?	CallOptions

Returns

Promise<Authorization>

Inherited from

AgentOperations.authorization

---

cancel()

cancel(options?): Promise<void>;

Stop what an agent is doing.

Parameters

Parameter	Type
options?	CallOptions

Returns

Promise<void>

Inherited from

AgentOperations.cancel

---

events()

events(input?, options?): Promise<Events>;

Read an agent's log.

Parameters

Parameter	Type	Description
input?	{ types?: string[]; after?: string; limit?: number; }	-
input.types?	string[]	The event types to read, as wardrik.run.completed, given once for each. Every type where left out. A wardrik.events.dropped comes whatever the types.
input.after?	string	The cursor to read after: a page's next, the id of an event a stream gave, or a snapshot's cursor.
input.limit?	number	How many to read, 100 unless set.
options?	CallOptions	-

Returns

Promise<Events>

Inherited from

AgentOperations.events

---

streamEvents()

streamEvents(input?, options?): EventStream;

Like events, then follows each new event as it's written. Starts after after, or at the end of the log. See EventStream.

Parameters

Parameter	Type	Description
input?	{ types?: string[]; after?: string; }	-
input.types?	string[]	The event types to read, as wardrik.run.completed, given once for each. Every type where left out. A wardrik.events.dropped comes whatever the types.
input.after?	string	The cursor to read after: a page's next, the id of an event a stream gave, or a snapshot's cursor.
options?	StreamOptions	-

Returns

EventStream

Inherited from

AgentOperations.streamEvents

---

timeline()

timeline(options?): Promise<Timeline>;

Read an agent's conversation as a person reads it.

Parameters

Parameter	Type
options?	CallOptions

Returns

Promise<Timeline>

Inherited from

AgentOperations.timeline

---

runs()

runs(options?): Promise<Runs>;

Read what an agent's runs did and cost.

Parameters

Parameter	Type
options?	CallOptions

Returns

Promise<Runs>

Inherited from

AgentOperations.runs

---

approve()

approve(
   interrupt,
   input?,
   options?
): Promise<object[]>;

Approve a call an agent waits on.

agent.decide with one approval. Recorded as the caller's.

Parameters

Parameter	Type
interrupt	string
input?	Approve
options?	CallOptions

Returns

Promise<object[]>

Inherited from

AgentOperations.approve

---

deny()

deny(
   interrupt,
   input?,
   options?
): Promise<object[]>;

Reject a call an agent waits on.

agent.decide with one rejection. Recorded as the caller's.

Parameters

Parameter	Type
interrupt	string
input?	Deny
options?	CallOptions

Returns

Promise<object[]>

Inherited from

AgentOperations.deny

---

definition()

definition(options?): Promise<DefinitionRead>;

Read an agent's definition.

Acts on the agent, never a conversation.

Parameters

Parameter	Type
options?	CallOptions

Returns

Promise<DefinitionRead>

Inherited from

AgentOperations.definition

---

spawn()

spawn(options?): Promise<AgentSummary>;

Start an agent.

Acts on the agent, never a conversation.

Parameters

Parameter	Type
options?	CallOptions

Returns

Promise<AgentSummary>

Inherited from

AgentOperations.spawn

---

stop()

stop(options?): Promise<AgentSummary>;

Stop an agent.

Parameters

Parameter	Type
options?	CallOptions

Returns

Promise<AgentSummary>

Inherited from

AgentOperations.stop

---

clear()

clear(options?): Promise<AgentSnapshot>;

Clear an agent's history.

Empties its history. The agent, its definition and its other conversations stay. Answers conflict while it runs a turn or waits on a person.

Parameters

Parameter	Type
options?	CallOptions

Returns

Promise<AgentSnapshot>

Inherited from

AgentOperations.clear

---

calls()

calls(options?): Promise<object[]>;

List the tool calls an agent is running.

Answers not_found while the agent sleeps.

Parameters

Parameter	Type
options?	CallOptions

Returns

Promise<object[]>

Inherited from

AgentOperations.calls

---

kill()

kill(call, options?): Promise<void>;

Stop one tool call.

Parameters

Parameter	Type
call	string
options?	CallOptions

Returns

Promise<void>

Inherited from

AgentOperations.kill

---

triggerSet()

triggerSet(
   trigger,
   input?,
   options?
): Promise<TriggerDefinition>;

Set one of an agent's triggers.

Sets or changes one trigger of an agent with a definition, which keeps it; not_found where it has none. A trigger whose every and timezone stay keeps its next run.

Acts on the agent, never a conversation.

Parameters

Parameter	Type
trigger	string
input?	TriggerDefinition
options?	CallOptions

Returns

Promise<TriggerDefinition>

Inherited from

AgentOperations.triggerSet

---

triggerRemove()

triggerRemove(trigger, options?): Promise<TriggerDefinition>;

Remove one of an agent's triggers.

Acts on the agent, never a conversation.

Parameters

Parameter	Type
trigger	string
options?	CallOptions

Returns

Promise<TriggerDefinition>

Inherited from

AgentOperations.triggerRemove

---

move()

move(input, options?): Promise<AgentSummary>;

Move an agent to another node.

Starts the agent again on node, as it was. A turn it's taking ends, and what it waits on is abandoned, so ask it again once it's moved.

Parameters

Parameter	Type
input	Node
options?	CallOptions

Returns

Promise<AgentSummary>

Inherited from

AgentOperations.move