golem-add-agent-scala

name	golem-add-agent-scala
description	Adding a new Scala agent to a Golem component. Use when the user asks to create, add, or define a new agent type, implement an agent trait, or add agent methods in a Scala Golem project.

Adding a New Agent to a Scala Golem Component

Overview

An agent is a durable, stateful unit of computation in Golem. Each agent type is defined as a trait + implementation class pair using annotations from golem.runtime.annotations.

Steps

1. Create the agent trait file — add src/main/scala/<package>/<AgentName>.scala
2. Create the agent implementation file — add src/main/scala/<package>/<AgentName>Impl.scala
3. Annotate the trait with @agentDefinition extending BaseAgent
4. Annotate the implementation with @agentImplementation()
5. Build — run golem build to verify

Agent Definition

The trait defines the agent's API:

import golem.runtime.annotations.agentDefinition
import golem.BaseAgent

import scala.concurrent.Future

@agentDefinition(mount = "/counters/{name}")
trait CounterAgent extends BaseAgent {

  class Id(val name: String)

  def increment(): Future[Int]

  def getCount(): Future[Int]
}

The implementation provides the behavior:

import golem.runtime.annotations.agentImplementation

import scala.concurrent.Future

@agentImplementation()
final class CounterAgentImpl(private val name: String) extends CounterAgent {
  private var count: Int = 0

  override def increment(): Future[Int] = Future.successful {
    count += 1
    count
  }

  override def getCount(): Future[Int] = Future.successful(count)
}

Agent Identity

The agent's constructor parameters define its identity. Declare them as an inner class Id(...) in the trait:

@agentDefinition()
trait ShardAgent extends BaseAgent {
  class Id(val region: String, val partition: Int)
  // ...
}

The implementation class takes the same parameters (as a tuple for multi-param constructors):

@agentImplementation()
final class ShardAgentImpl(input: (String, Int)) extends ShardAgent {
  private val (region, partition) = input
  // ...
}

Custom Types

Use case classes for structured data. The SDK requires a zio.blocks.schema.Schema for custom types used as method parameters or return values. For collections, use List[T] instead of Array[T] — Array does not have automatic Schema derivation support:

import zio.blocks.schema.Schema

final case class Coordinates(lat: Double, lon: Double) derives Schema
final case class WeatherReport(temperature: Double, description: String) derives Schema

@agentDefinition()
trait WeatherAgent extends BaseAgent {
  class Id(val apiKey: String)
  def getWeather(coords: Coordinates): Future[WeatherReport]
}

Returning Failures

Agent methods should distinguish between domain errors (expected failure outcomes) and uncaught errors:

• Uncaught errors (Future.failed, thrown exceptions that escape the method) are not returned to the caller as a failed invocation. Golem treats them as crashes: the invocation is retried according to the agent's retry policy, and if the retries are exhausted the agent itself becomes failed.
• Domain errors that the caller should observe as a normal failure result must be expressed in the method's return type using Either[E, A] wrapped in a Future. Both E and A need a Schema instance.

import golem.runtime.annotations.{agentDefinition, agentImplementation}
import golem.BaseAgent
import zio.blocks.schema.Schema

import scala.concurrent.Future

enum WithdrawError derives Schema:
  case InsufficientFunds(available: Long)
  case AccountClosed

@agentDefinition(mount = "/wallets/{owner}")
trait Wallet extends BaseAgent {
  class Id(val owner: String)

  def withdraw(amount: Long): Future[Either[WithdrawError, Long]]
}

@agentImplementation()
final class WalletImpl(private val owner: String) extends Wallet {
  private var balance: Long = 0L
  private var closed: Boolean = false

  override def withdraw(amount: Long): Future[Either[WithdrawError, Long]] =
    Future.successful {
      if (closed) Left(WithdrawError.AccountClosed)
      else if (amount > balance) Left(WithdrawError.InsufficientFunds(balance))
      else {
        balance -= amount
        Right(balance)
      }
    }
}

Returning Left(WithdrawError.X) completes the invocation successfully — the caller receives the error as a value. Returning a failed Future (e.g. Future.failed(new RuntimeException(...))) or letting an exception escape will instead trigger a retry and eventually fail the whole agent.

HTTP API Annotations

Agents can expose methods as HTTP endpoints using @endpoint and @header:

import golem.runtime.annotations.{endpoint, header}

@agentDefinition(mount = "/api/{id}")
trait ApiAgent extends BaseAgent {
  class Id(val id: String)

  @endpoint(method = "GET", path = "/data")
  def getData(@header("Authorization") auth: String): Future[String]

  @endpoint(method = "POST", path = "/update")
  def update(body: UpdateRequest): Future[UpdateResponse]
}

Related Skills

• Load golem-js-runtime for details on the QuickJS runtime environment, available Web/Node.js APIs, and npm compatibility
• Load golem-file-io-scala for reading and writing files from agent code

Key Constraints

• All agent traits must extend BaseAgent and be annotated with @agentDefinition
• All agent implementations must be annotated with @agentImplementation()
• Custom types used in agent methods require a zio.blocks.schema.Schema instance (use derives Schema in Scala 3)
• Constructor parameters define agent identity — they must be serializable types with Schema instances
• The class Id(...) inner class in the agent trait defines the constructor parameter schema
• The implementation takes constructor params directly (single param) or as a tuple (multi-param)
• Agents are created implicitly on first invocation — no separate creation step
• Invocations are processed sequentially in a single thread — no concurrency within a single agent
• The scalacOptions += "-experimental" flag is required for macro annotations