Memory indexer silently fails — 0 chunks indexed, exit code 1

Version: 2026.2.25 and 2026.2.26
OS: macOS Sonoma, Intel i5 (MacBook Pro 2013), 8GB RAM
Node: v22.22.0

Problem

openclaw memory index runs through all files, logs "embeddings: batch start" for every batch, then exits with code 1. Zero chunks are written to the SQLite store. No error message is logged anywhere (gateway.log, gateway.err.log, or CLI output).

Tested with 3 different providers — same result:

1. provider: "gemini" — embeddings ready, 0 chunks
2. provider: "openai" (Ollama at localhost:11434/v1/) — embeddings ready, manual curl works fine (768-dim vectors returned), 0 chunks
3. provider: "local" (node-llama-cpp, embeddinggemma-300m) — model downloaded, embeddings ready, 0 chunks

Reproduction

rm -f ~/.openclaw/memory/main.sqlite
openclaw memory index --agent main --verbose

Output:

[memory] sync: indexing memory files
[memory] embeddings: batch start
[memory] embeddings: batch start
... (repeated ~36 times for memory, ~22 times for sessions)
[memory] sync: indexing session files
[memory] embeddings: batch start
... (repeated)
# exits with code 1, no error

openclaw memory status --deep --agent main

Shows: Indexed: 0/288 files · 0 chunks, Dirty: yes, Embeddings: ready, Vector: ready, FTS: ready

What works

• Embedding providers connect and respond correctly
• Manual embedding requests return valid vectors
• sqlite-vec extension loads fine
• Memory status reports everything as "ready"

What's broken

• Batches start but never complete/write
• No error is surfaced — silent failure
• Exit code 1 with zero diagnostic output

Config (relevant parts)

{
  "memorySearch": {
    "provider": "local",
    "fallback": "none",
    "sources": ["memory", "sessions"],
    "experimental": { "sessionMemory": true },
    "query": {
      "hybrid": {
        "enabled": true,
        "vectorWeight": 0.7,
        "textWeight": 0.3,
        "mmr": { "enabled": true, "lambda": 0.7 },
        "temporalDecay": { "enabled": true, "halfLifeDays": 30 }
      }
    }
  }
}

Files: 33 memory markdown files, 255 session JSONL files.