The refreshed provider table said 12 embedders; EmbedderFactory registers 11.
The other three rows were right, so the table drifted silently again inside the
commit that was fixing it drifting.
Add tests/test_agents_md_provider_counts.py, which parses the AGENTS.md provider
table and asserts each documented count equals len(<Factory>.provider_to_class).
Verified it fails on the 12 and passes on 11.
Commit a488e1904 deleted mem0/graphs/ and mem0-ts/src/oss/src/graphs/ when the
v3 pipeline landed, and 3f717e545 reframed the docs as Platform-only, but the
agent-facing context files were never updated. Agents reading them still
believe OSS graph memory exists and propose graph_store config that cannot work.
- AGENTS.md: drop graphs/ from the directory map and dependency tree, correct
the provider table (4 categories, not 5; LLMs 24->18, vector stores 30->25,
embeddings 15->12), and rewrite the Graph Memory section as Platform-only
- skills: replace the OSS graph-memory prompt example and stop indexing
/open-source/features/graph-memory, which 404s; also fix four other
SECTION_MAP paths that no longer resolve
- mem0/exceptions.py: DependencyError example referenced kuzu/graph_store
- tests/test_telemetry.py: drop four mock_memory.config.graph_store lines;
graph_store is not a MemoryConfig field and telemetry never reads it
- CONTRIBUTING.md: add the three missing release tag prefixes (opencode-v*,
pi-agent-v*, n8n-nodes-mem0-v*) and note Zapier deploys off-registry
kartik-mem0