Documentation
¶
Overview ¶
Path B — LangGraph graph nodes for automatic memory injection, mirroring the Python langgraph/nodes.py on the main branch. Wire the returned NodeFuncs into the host's own StateGraph (graph.AddNode) over a state with a "messages" channel (channels.MessagesReducer, []messages.Message).
Shared runtime helpers for the LangGraph integration (paths A and B), mirroring Python's langgraph/_runtime.py on the main branch. Keeps Engine lifecycle and workspace resolution in one place so the two paths stay consistent and have a single seam against Engine.
Package langgraph integrates ladyM as a memory layer for langchain-golang LangGraph hosts — the Go port of the Python langgraph module (main branch).
Two paths, mirroring the Python module:
- Path A (Tools): CreateTools returns LangChain tools wrapping ladyM long-term memory; the LLM decides when to recall / remember (ReAct-style, e.g. via agents.CreateAgent).
- Path B (Nodes): CreateRecallNode / CreateRetainNode return graph NodeFuncs for automatic per-turn memory injection, wired into the host's own StateGraph.
Index ¶
- func CreateRecallNode(eng *engine.Engine, topK int, prefix string, wsFn WorkspaceFunc) lcgraph.NodeFunc
- func CreateRetainNode(eng *engine.Engine, wsFn WorkspaceFunc) lcgraph.NodeFunc
- func CreateTools(eng *engine.Engine, workspace string, defaultTopK int) ([]lctools.Tool, error)
- func ResolveEngine(src any, workspace string) (*engine.Engine, error)
- type WorkspaceFunc
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func CreateRecallNode ¶
func CreateRecallNode(eng *engine.Engine, topK int, prefix string, wsFn WorkspaceFunc) lcgraph.NodeFunc
CreateRecallNode returns a graph node that recalls against the latest message and prepends a SystemMessage to state["messages"], mirroring Python's create_recall_node. Emits no update when there are no hits.
func CreateRetainNode ¶
func CreateRetainNode(eng *engine.Engine, wsFn WorkspaceFunc) lcgraph.NodeFunc
CreateRetainNode returns a graph node that stores the latest human+AI turn into long-term memory (subject to ladyM's attention gate), mirroring Python's create_retain_node. For per-request multi-user isolation (a resolved workspace different from the engine default) a short-lived workspace-bound Engine is built per call and closed afterwards.
func CreateTools ¶
CreateTools builds LangChain tools backed by ladyM memory: [recall_memory, remember_fact, search_code]. Each tool closes over the given Engine; workspace overrides the engine's default workspace (empty keeps the engine default). Mirrors Python's create_ladym_tools.
func ResolveEngine ¶
ResolveEngine returns a ready-to-use Engine from flexible factory input, mirroring Python's resolve_engine:
- *engine.Engine — returned as-is (caller owns lifecycle); workspace is IGNORED — the engine's own Config.Workspace wins. Pass a *config.Config or db path instead if you need a different workspace.
- *config.Config — new Engine from a shallow copy; a non-empty workspace overrides cfg.Workspace.
- string — treated as a db path over config.Default(); workspace honored.
- nil — config.Default() (offline hashing embedding); workspace honored.
Any other type is an error.
Types ¶
type WorkspaceFunc ¶
WorkspaceFunc resolves the ladyM workspace for a node run, mirroring Python's per-request isolation via config["configurable"]["user_id"]. A nil WorkspaceFunc uses the engine's default workspace.
func WorkspaceFromUserID ¶
func WorkspaceFromUserID() WorkspaceFunc
WorkspaceFromUserID returns a WorkspaceFunc that resolves the ladyM workspace from the run-scoped context's "user_id" field, mirroring Python's resolve_workspace reading config["configurable"]["user_id"]. An empty result falls back to the engine's default workspace.
Source Files