Gralkor.AgentLearning (jido_gralkor v4.1.0)

One flat experiential-learning record: what kind of problem was approached, the approach taken, whether it succeeded, and the lesson learned.

No ERL-internal edges — it is a single node so ERL recall is single-label and never traverses. to_episode/1 renders it into a well-formed episode body graphiti ingests into the same group_id as ordinary memory; the body states the problem_kind and the outcome so a problem-kind-seeded hybrid search surfaces it, with the success bias living in the text. Any domain entities the lesson mentions are linked to the consumer's Learning node by graphiti.

See ex-agent-learning in TEST_TREES.md.

Summary

Types

t()

Functions

to_episode(learning)

Render the learning into the episode body graphiti ingests and recalls by problem kind.

Types

t()

@type t() :: %Gralkor.AgentLearning{
  approach: String.t(),
  lesson: String.t(),
  problem_kind: String.t(),
  success: boolean()
}

Functions

to_episode(learning)

@spec to_episode(t()) :: String.t()

Render the learning into the episode body graphiti ingests and recalls by problem kind.