feat(memory): add LangMemMemoryService with semantic and episodic extraction

[miyannishar]

Summary

Adds LangMemMemoryService — a BaseMemoryService implementation backed by LangMem and any LangGraph BaseStore, providing self-hosted long-term memory with semantic search for ADK agents.

This addresses the long-term memory portion of #46 (Tier 2: persistent facts, semantic search, cross-session persistence). The session service portion (Tier 1: working memory with auto-summarization and context window management) is not included here — that would require wrapping the Redis Agent Memory Server client and is a separate effort.

Changes

• src/google/adk_community/memory/langmem_memory_service.py — full implementation
• tests/unittests/memory/test_langmem_memory_service.py — 21 unit tests with async mock infrastructure
• src/google/adk_community/memory/__init__.py — export LangMemMemoryService
• pyproject.toml — add langmem optional dependency group

Design Decisions

Concern	Approach
Backend	Accepts any LangGraph BaseStore — InMemoryStore for dev, AsyncPostgresStore for production
Extraction model	User-provided model string (e.g. "openai:gpt-4o-mini", "google:gemini-2.0-flash")
Memory tiers	Dual-tier: semantic (user facts/preferences) + episodic (tool call narratives)
Episodic opt-out	enable_episodic=False skips episodic extraction for simpler use cases
Namespacing	(*prefix, user_id, "semantic"|"episodic") — fully configurable prefix
Sensitive data	Default extraction prompts instruct LLM to omit credentials
Error handling	Partial failures logged, not raised — one tier failing doesn't block the other
Optional dependency	langmem + langgraph only required via pip install google-adk-community[langmem]

Usage

fromlanggraph.store.memoryimportInMemoryStorefromgoogle.adk_community.memoryimportLangMemMemoryServicestore=InMemoryStore(
index={"dims": 1536, "embed": "openai:text-embedding-3-small"}
)
memory_service=LangMemMemoryService(
store=store,
extraction_model="openai:gpt-4o-mini",
)
# Use with ADK Runnerrunner=Runner(
app_name="my_app",
agent=agent,
memory_service=memory_service,
)

Test Plan

pip install google-adk-community[langmem,test]
pytest tests/unittests/memory/ -v

• 21 unit tests covering extraction, search, namespace scoping, error handling, and edge cases
• All tests pass without LangMem installed (full mock infrastructure)
• Existing memory tests unaffected

Related: #46 (long-term memory only — session service is a separate effort)

[DeanChensj]

@gemini-cli /review

[github-actions]

🤖 Hi @DeanChensj, I've received your request, and I'm working on it now! You can track my progress in the logs for more details.

[github-actions]

Thanks for the PR! This is a solid implementation of LangMemMemoryService. The dual-tier approach (semantic + episodic) is well-designed, and the test coverage is impressive.

I've left a few minor suggestions regarding robustness in message extraction and search result sorting. Most of these are edge-case considerations.

Overall, great work!