memx

Adapters

--adapter is a built-in short name or an import path module.path:ClassName. Implement export_state and (for async stores) wait_until_ready so diagnostics can snapshot the store; if export/query are inconsistent, Stage 1 and Stage 4 false-positive.

Name Class Extra
mem0 memx.adapters.mem0:Mem0Adapter pip install -e ".[mem0]"
supermemory memx.adapters.supermemory:SupermemoryAdapter pip install -e ".[supermemory]" 
pip install -e ".[providers]"   # mem0 + supermemory

Mem0

  • MEM0_API_KEY set: Mem0 platform client.
  • Unset: OSS Memory() and OPENAI_API_KEY. Extraction LLM defaults to gpt-5-mini; override with MEM0_LLM_MODEL.
  • Ingest uses async_mode when the SDK accepts it. wait_until_ready polls event IDs in parallel.
  • GPT-5 family models reject temperature=0.1; the adapter marks them as reasoning models so Mem0 does not send that parameter.

SuperMemory

  • SUPERMEMORY_API_KEY required unless SUPERMEMORY_BASE_URL points at a keyless self-host.
  • Ingest: dreaming="instant", custom_id=session_id, container_tag=entity_id.
  • wait_until_ready polls documents.get. Ready when dreaming_status is done (memories exist). Document status=indexing after that is not treated as blocking.
  • Query: hybrid search (search_mode=hybrid, threshold 0.3) over memories and chunks.
  • export_state is profile() (static + dynamic). Index-based fact ids mean diffs can look like mass remove/add even when content is similar.
  • reset(entity_id) calls documents.delete_bulk by container tag. memx run never calls reset.

Custom adapter

Subclass BaseMemoryAdapter and pass the class on the CLI:

memx run --dataset locomo --adapter mypkg.adapters:MyAdapter --limit 3 --random

The class must be constructible with no required arguments (env vars / defaults). Implement ingest_session, query, export_state. Override wait_until_ready if ingest returns before the store is queryable. See python-api.md.

Ingest scheduling

Context sessions with no selected questions are ingested concurrently, then one wait_until_ready covers that batch. A session that has selected questions is ingested and waited on alone so the state diff for diagnostics belongs to that session. Independent conversations (case_id / LoCoMo sample_id) run with --concurrency.