Modules
In-flight conversation buffer keyed by session_id.
Public entry point and adapter port for Gralkor memory.
In-memory twin of Gralkor.Client for tests.
Production Gralkor.Client implementation. In-process — no HTTP — talks
to graphiti via Gralkor.GraphitiPool (Pythonx-backed) and to the LLM via
req_llm (Elixir-side, used by Gralkor.Distill and Gralkor.Interpret).
Configuration for the embedded Gralkor runtime.
A named memory destination with an address and extraction ontology.
Render a list of conversation turns into the transcript episode body ingested into the knowledge graph.
Format graphiti edge data into the canonical fact strings the LLM sees.
A complete graph payload supplied to Gralkor.Client.replace/1.
Per-group Graphiti instance cache, plus the gateway for graphiti operations.
A request to ingest content through a registered Gralkor.Lens.
The successful result of storing one piece of evidence through one Lens.
Filter retrieved graph facts down to those relevant to the conversation, using the configured LLM.
Raised when the LLM's interpret response cannot be parsed against the
structured-output schema — typically because :output_token_budget was too
small and the model's answer truncated mid-list, but also for any other
schema mismatch.
Application-owned definition of a memory Lens.
Behaviour for consumer-defined Lens ingestion processes.
Resolved definition of a Lens whose write unit is a complete graph.
Storage capability passed to a Gralkor.Lens.Ingestion process.
Canonical message shape that Gralkor expects at its port boundary.
Declare an entity-and-relationship ontology for graphiti's custom entity extraction.
PythonX runtime owner for the embedded Gralkor stack.
Orchestrate one recall call: search the graph, interpret the hits against
the buffered conversation, wrap the result in a <gralkor-memory> block.
A named post-ingestion process that saves to a Destination.
One structured result produced by one completed Reflection run.
Runs a repository-defined Chain of Thought as an ordered, tool-capable inference sequence.
Schedules independent Reflections once after a multi-Lens ingestion has completed.
Stores and searches artefacts in destinations named by Reflections.
A request to replace all graph content owned by one replaceable Lens.
A search across one or more registered memory Destinations.
ReAct tool the LLM can call to store a thought or decision in memory.
Admin tool that runs Graphiti's community detection over this agent's memory group.
Admin tool that rebuilds Gralkor's graph search indices.
ReAct tool the LLM can call to search long-term memory.
Converts a Jido/ReAct turn into Gralkor's canonical message shape.
Memory consolidation for a long-lived Jido agent.
Jido.AgentServer.Lifecycle implementation that flushes the active
Gralkor session when an agent process terminates.
Jido plugin that replaces Jido.Memory.BasicPlugin with Gralkor-backed
memory. Claims the :__memory__ slot so it is the only memory plugin
attached to the agent.
Helpers for Jido.AI.Reasoning.ReAct consumers that want Gralkor-aware
request shaping.