Objectives

Every run.
In the open.

Follow every message and tool call. Debug the details, compact long conversations, and control the work through the TypeScript SDK.

Get started

Start an objective

const objective = await cadenya.objectives.create({
  agentId,
  firstUserMessage: "Update the customer on order 1042"
});

Follow it live

const events = await cadenya.objectives.streamEvents(objective.metadata.id);
for await (const event of events) {
  console.log(event.data);
}

Every step, in order.

Messages, tool calls, results, and status changes share one timeline. Stream events live or fetch the history from your application.

Timeline

User message
lookup_order 180 ms
get_shipment 2.0 s
get_shipment 420 ms
Approval received
send_customer_update 310 ms
Assistant message 723 ms

Read the run and its events

const objective = await cadenya.objectives
  .retrieve(objectiveId);
const events = await cadenya.objectives
  .listEvents(objectiveId, { sinceEventId });

Inspect every tool call.

Inspect arguments, execution status, and returned content. Approve or deny gated calls through the TypeScript SDK.

send_customer_update

Approved · Completed

Arguments

{ "orderId": "1042",
  "channel": "email" }

Result

{ "delivered": true }

Inspect a call. Approve it.

const call = await cadenya.objectives
  .retrieveToolCall(objectiveId, toolCallId);

await cadenya.objectives
  .approveToolCall(objectiveId, toolCallId);

Track every
tool call
your agent makes

Get past the error message.

Trace requests, responses, timing, and attempts in the debugger. Read context composition and token usage through the diagnostics API.

get_shipment

Attempt 1 · 2.0 s

GET /shipments/SHP-1042

{ "shipmentId": "SHP-1042" }

HTTP 504 · Gateway timeout

UPSTREAM_TIMEOUT
Carrier request timed out.

Inspect context usage

const { diagnostics } = await cadenya.objectives
  .retrieveDiagnostics(objectiveId);

Input tokens: 1,529
Cached input: 1,408

Keep long tasks moving.

Compaction runs automatically as context fills. Summarize the conversation, clear older tool results, or combine both. Inspect each window or compact on demand.

Context windows

Automatic · 75% trigger

Window 01

Messages + tool results

Compaction

Summary / result clearing

Window 02

Order 1042 + shipment context

Compact. Keep going.

await cadenya.objectives.compact(objectiveId);

const windows = await cadenya.objectives
  .listContextWindows(objectiveId);

A few things
to know.

What does an objective keep?

An objective captures its configuration at creation and records its messages, tool calls, events, and context windows.

Can I send another instruction?

Yes. Continue a waiting objective, or enqueue a message while its agent loop is busy.

Does compaction happen automatically?

Yes. The default trigger is 75% of the model’s context limit. Configure summarization, older-result clearing, or both on the variation.

Give agents
an objective.

Start a run. Stream its events. Keep control as it works.

Get started See API docs