PURISTA

Appearance

Runtime ​

An agent definition (AgentBuilder) is inert. To run it, you must create an instance and provide it with concrete runtime dependencies (Model Providers, Databases, etc.).

1. Bootstrapping the Instance ​

getInstance(eventBridge, options) is where you inject your production-ready tools.

ts
const supportAgentInstance = await supportAgent.getInstance(eventBridge, {
  models: {
    'openai:gpt-4o-mini': new AiSdkProvider({ model: openai('gpt-4o-mini') })
  },
  conversationStore: new RedisConversationStore(),
  knowledgeAdapters: {
    supportFaq: new VectorStoreAdapter()
  },
  poolConfig: {
    poolId: 'support',
    maxConcurrencyPerInstance: 5
  }
})

await supportAgentInstance.start()

2. Managing Concurrency (Pools) ​

LLM calls are expensive and can be slow. To protect your application and manage rate limits, PURISTA uses Worker Pools.

  • poolId: Groups multiple agent instances into a shared concurrency limit.
  • maxConcurrencyPerInstance: Limits how many agent runs can happen in parallel within a single process.

Why use Pools? ​

  • Avoid Resource Exhaustion: Prevents one agent from hogging all event-loop resources or memory.
  • Rate Limit Protection: Keeps your outgoing LLM requests within your provider's quota.
  • Fairness: Ensures that high-priority agents still have "slots" to run even during peak traffic.

3. Deployment Patterns ​

Pattern A: In-Process (Monolith/Service) ​

Run the agent in the same process as your API or Service. Good for low-to-medium volume or real-time streaming needs.

Pattern B: Isolated Workers (Microservice) ​

Deploy a dedicated process that only runs agents. This allows you to scale AI workloads independently from your web traffic.

4. Health & Monitoring ​

Every agent instance provides a read-only status snapshot:

ts
const status = supportAgentInstance.getStatus()
/*
{
  poolId: 'support',
  activeWorkers: 2,
  waitingWorkers: 1,
  maxConcurrencyPerInstance: 5
}
*/

This data is automatically included in PURISTA's health checks and telemetry frames, allowing you to alert on pool congestion.