Bug

When memory.backend = "qmd" is set, the memory_search tool still tries to resolve cloud embedding providers (OpenAI/Google/Voyage) instead of routing through QMD's local search.

Config

"memory": { "backend": "qmd" },
"memorySearch": { "provider": "local", "fallback": "none" }

Error

disabled: true,
error: "No API key found for provider openai... google... voyage..."

Expected

When memory.backend = "qmd", memory_search should delegate to qmd query (which works perfectly with local embeddings). The builtin embedding provider resolution should be skipped entirely.

Workaround

Calling qmd query directly via exec works fine. The issue is only in the tool wrapper.

Version

2026.2.6-3