Oban worker for durable agent execution.

Runs ADK agents as background jobs with retries, scheduling, and persistence.

Setup

1. Add oban to your deps
2. Configure Oban in your application (see Oban docs)
3. Enqueue jobs using new/2 or the helper enqueue/4

Usage

# Direct Oban usage
%{
  agent_module: "MyApp.Agents.Helper",
  user_id: "user1",
  session_id: "sess1",
  message: "Hello!",
  app_name: "my_app"
}
|> ADK.Oban.AgentWorker.new(queue: :agents, max_attempts: 5)
|> Oban.insert()

# Helper function
ADK.Oban.AgentWorker.enqueue(
  MyApp.Agents.Helper,
  "user1",
  "Hello!",
  app_name: "my_app",
  session_id: "custom-session",
  queue: :agents,
  priority: 1
)

Agent Resolution

The worker resolves agents via agent_module — a module that implements a agent/0 callback returning an ADK.Agent.t() struct:

defmodule MyApp.Agents.Helper do
  def agent do
    ADK.Agent.LlmAgent.new(
      name: "helper",
      model: "gemini-flash-latest",
      instruction: "You are a helpful assistant."
    )
  end
end

Alternatively, pass agent_config as a map with type and agent params for inline agent construction (useful for simple cases):

%{
  agent_config: %{
    "type" => "llm",
    "name" => "helper",
    "model" => "gemini-flash-latest",
    "instruction" => "Be helpful"
  },
  user_id: "user1",
  message: "Hello!"
}
|> ADK.Oban.AgentWorker.new()
|> Oban.insert()

Result Storage

By default, results are broadcast via ADK.Telemetry:

:telemetry.execute(
  [:adk, :oban, :job, :complete],
  %{duration: duration_ms},
  %{job_id: id, events: events, args: args}
)

Attach a telemetry handler to persist results as needed.

Configuration

config :adk, ADK.Oban.AgentWorker,
  default_queue: :agents,
  default_max_attempts: 3,
  default_priority: 2