Reference

Agents

pipeline, session, prompt, and guard blocks for agent orchestration.

Summary

Agent blocks orchestrate multi-step LLM flows, conversational sessions, versioned prompts, and scoped file writes — checked like any other Point source.

pipeline

Multi-step flows with typed events. See examples/pipelines/document-ingest.point for step ... is await action(...) with retry and policy guards.

session

Conversational state with streaming actions. See examples/agents/support-chat.point.

prompt

Versioned templates with record placeholders. See examples/prompts/support-greeting.point:

module SupportGreeting

record Support Context
  user name: Text
  tier: Text

prompt support greeting
  version 1
  input Support Context
  template "Hello {user name}, thank you for contacting support. Your {tier} plan is ready to help."

guard output paths

Scope file writes in pipelines. See examples/pipelines/guarded-output.point:

module GuardedOutput

guard output paths
  allow "output/**"
  allow "tmp/*"

std.ai providers

Use std.ai for OpenAI and Anthropic provider actions. See AI provider interop.

See also