Bug Description

When running the CLI command openclaw memory index --agent devops, it fails with the error:

Memory index failed (devops): Unknown memory embedding provider: ollama

The Gateway runtime works correctly with the same configuration, but the CLI commands fail to initialize the Ollama embedding provider.

Environment

• OpenClaw Version: 2026.4.9 (0512059)
• OS: Linux (x86_64)
• Node.js: v22.21.1
• Memory Provider: ollama
• Embedding Model: qwen3-embedding:0.6b

Configuration

Relevant openclaw.json settings:

"memorySearch": {
  "enabled": true,
  "provider": "ollama",
  "sync": {
    "watch": true,
    "sessions": {
      "deltaBytes": 100000,
      "deltaMessages": 50
    }
  },
  "model": "qwen3-embedding:0.6b"
}

Steps To Reproduce

1. Configure memorySearch.provider as "ollama" and memorySearch.model as "qwen3-embedding:0.6b"
2. Ensure Ollama is running and the model is available (ollama list shows the model)
3. Run: openclaw memory index --agent devops
4. CLI fails with: Unknown memory embedding provider: ollama

Expected Behavior

The CLI should successfully initialize the Ollama embedding provider and index the memory files.

Actual Behavior

17:06:31+08:00 [plugins] [lcm] Plugin loaded
🦞 OpenClaw 2026.4.9 (0512059)

│
◇  
Memory index failed (devops): Unknown memory embedding provider: ollama
Error: (none)
Exit Code: 1

Additional Context

What Works

• ✅ Ollama is running and responsive
• ✅ Model qwen3-embedding:0.6b exists and works:

  $ curl -s http://localhost:11434/api/embed -d '{"model": "qwen3-embedding:0.6b", "input": "test"}'
  # Returns embeddings successfully

• ✅ Gateway runtime works correctly (memory indexing via Gateway succeeds)
• ✅ Configuration is valid (Gateway loads it without errors)

What Fails

• ❌ All CLI memory commands fail:
  • openclaw memory index --agent devops
  • openclaw memory status --agent devops
  • openclaw memory search "query"

Stack Trace

Error: Unknown memory embedding provider: ollama
    at getAdapter (file:///root/.nvm/versions/node/v22.21.1/lib/node_modules/openclaw/dist/manager-vTWIfqhx.js:317:22)
    at createEmbeddingProvider (file:///root/.nvm/versions/node/v22.21.1/lib/node_modules/openclaw/dist/manager-vTWIfqhx.js:369:25)
    at MemoryIndexManager.loadProviderResult (...)

Root Cause Analysis

The builtin embedding providers (including ollama) are registered via registerBuiltInMemoryEmbeddingProviders(api), which is called in the memory-core plugin's register() function.

The Gateway loads plugins through the full lifecycle (which calls register()), but the CLI does not trigger this registration path. As a result, the provider registry is empty when CLI commands attempt to resolve the embedding provider.

This is consistent with the issue observed in #29568 and #10931, but specifically affects CLI commands rather than the memory_search tool.

Workaround

Currently, there is no CLI workaround. Users must rely on the Gateway's automatic file watcher (sync.watch: true) to index memory files.

Suggested Fix

Ensure the CLI runtime calls registerBuiltInMemoryEmbeddingProviders() before attempting to create embedding providers, or initialize the provider registry in the CLI entry point.