Unknown memory embedding provider: ollama - Troubleshooting Guide

Error Manifestation

When executing the openclaw memory status --deep command with ollama configured as the memory embedding provider, the CLI terminates immediately with a provider resolution failure:

$ openclaw memory status --deep

🦞 OpenClaw 2026.4.12 (1c0672b) — If something's on fire, I can't extinguish it—but I can write a beautiful postmortem.

│
◇  
[openclaw] Failed to start CLI: Error: Unknown memory embedding provider: ollama

Technical Manifestations

• Exit code: Non-zero (typically 1)
• Error type: ProviderResolutionError or equivalent in the memory subsystem
• Error message: Unknown memory embedding provider: ollama
• Stack trace location: Likely in packages/core/src/memory/providers/registry.ts or equivalent

Affected Configurations

The error occurs when the OpenClaw configuration contains:

{
  "memory": {
    "embedding": {
      "provider": "ollama",
      "model": "qwen3-embedding:0.6b"
    }
  }
}

Or via environment variable:

$ export OPENCLAW_MEMORY_EMBEDDING_PROVIDER=ollama
$ export OPENCLAW_MEMORY_EMBEDDING_MODEL=qwen3-embedding:0.6b

Regression Timeline

• Last working version: 2026.04.10
• First failing version: 2026.04.12
• Version delta: 2 days of commits

Primary Root Cause: Provider Registry Regression

The error originates from a regression in the memory embedding provider registry system. The ollama provider was either:

1. Removed from the provider registry during a refactoring commit between 2026.04.10 and 2026.04.12
2. Renamed without backward compatibility (e.g., "ollama" → "ollama-embedding" or "ollama/text-embedding-2")
3. Excluded from the build bundle due to a tree-shaking issue or conditional import
4. Registering conditionally based on a feature flag that defaults to disabled

Code Flow Analysis

The failure sequence follows this path:

CLI Entry (memory/status.ts)
  → MemoryService.initialize()
    → EmbeddingProviderFactory.resolve("ollama")
      → ProviderRegistry.get("ollama")
        → ❌ throws "Unknown memory embedding provider: ollama"

Likely Commit Pattern

Based on the regression window, a refactoring commit likely changed the provider registration mechanism:

Before (working):

// packages/core/src/memory/providers/index.ts
export { OllamaEmbeddingProvider } from './ollama';
// Auto-registers via static side effect or explicit registry call

After (broken):

// packages/core/src/memory/providers/index.ts
// OllamaEmbeddingProvider export removed or conditional
// Provider not auto-registering

Alternative Root Cause: Configuration Schema Change

The provider name may have changed in the configuration schema:

// Old config key (2026.04.10)
memory.embedding.provider = "ollama"

// New config key (2026.04.12)
memory.embedding.provider = "ollama-embed"  // or
memory.embedding.provider = "ollama/text-embedding-2"

Verification Command

To diagnose the root cause, inspect the available providers:

$ openclaw memory providers list
# or
$ cat ~/.openclaw/config.json | jq '.memory.embedding'

Option 1: Revert to Previous Provider Name (Quick Fix)

If the provider was renamed, use the new configuration value:

# Check current OpenClaw version
openclaw --version

# List available memory embedding providers
openclaw memory status --help 2>&1 | grep -i provider

# Update configuration to use correct provider name
openclaw config set memory.embedding.provider ollama-embedding
# OR
openclaw config set memory.embedding.provider ollama/text-embedding-2

Option 2: Force Re-registration via Plugin (Workaround)

If the provider code exists but isn’t auto-registering:

# Create a local plugin to re-register the provider
mkdir -p ~/.openclaw/plugins/ollama-fix
cd ~/.openclaw/plugins/ollama-fix

cat > plugin.ts << 'EOF'
import { registerEmbeddingProvider } from '@openclaw/core';

export function registerOllamaProvider() {
  registerEmbeddingProvider('ollama', {
    name: 'ollama',
    createClient: () => new OllamaEmbeddingClient()
  });
}
EOF

# Enable plugin in config
openclaw config set plugins.enabled "['ollama-fix']"

Option 3: Reinstall/OpenClaw (Clean Fix)

Uninstall and reinstall to get a known-good configuration:

# Backup current configuration
cp -r ~/.openclaw ~/.openclaw.backup.$(date +%Y%m%d)

# Reinstall OpenClaw
npm uninstall -g @openclaw/cli
npm install -g @openclaw/cli@latest

# Reconfigure ollama provider
openclaw config set memory.embedding.provider ollama
openclaw config set memory.embedding.model qwen3-embedding:0.6b

Option 4: Pin to Working Version (Temporary Fix)

If the regression blocks production usage:

# Uninstall current version
npm uninstall -g @openclaw/cli

# Install last known working version
npm install -g @openclaw/[email protected]

# Verify version
openclaw --version

Configuration File Manual Edit

If CLI commands fail, directly edit the configuration:

# Edit the configuration file
nano ~/.openclaw/config.json

# Ensure memory section has correct provider:
{
  "memory": {
    "embedding": {
      "provider": "ollama",
      "model": "qwen3-embedding:0.6b"
    }
  }
}

Verify Fix: Basic Status Check

After applying the fix, verify with the original command:

$ openclaw memory status --deep

🦞 OpenClaw 2026.4.12 (1c0672b) — If something's on fire, I can't extinguish it—but I can write a beautiful postmortem.

│
◇  
Memory Search (main)
Provider: ollama (requested: ollama)
Model: qwen3-embedding:0.6b
Sources: memory
Indexed: 8/8 files · 99 chunks
Dirty: yes
Store: ~/.openclaw/memory/main.sqlite
Workspace: ~/.openclaw/workspace
Dreaming: 0 3 * * * · limit=10 · minScore=0.8 · minRecallCount=3 · minUniqueQueries=3 · recencyHalfLifeDays=14 · maxAgeDays=30
Embeddings: ready

Expected output:

• Exit code: 0
• No error messages
• Provider: ollama displayed correctly
• Embeddings: ready status

Verify Fix: Direct Embedding Test

Test the embedding functionality directly:

$ openclaw memory embed --text "test query"

# Expected: Returns embedding vector without error
# Exit code: 0

Verify Fix: Provider Registration (Debug)

If still failing, check provider registration:

$ openclaw debug providers

Available Memory Providers:
- openai
- anthropic
- ollama          ← Should be listed
- local

Verify Fix: Version Confirmation

$ openclaw --version

# Verify version matches expected
🦞 OpenClaw 2026.4.12 (1c0672b)

Verify Fix: Ollama Service

Ensure the Ollama service is running and accessible:

$ curl http://localhost:11434/api/tags

# Expected: JSON response with available models
{
  "models": [
    {
      "name": "qwen3-embedding:0.6b",
      "size": 378456789,
      "modified_at": "2026-04-10T00:00:00Z"
    }
  ]
}

Logically Connected Error Codes

• UNKNOWN_PROVIDER - Generic provider resolution failure
• PROVIDER_NOT_INITIALIZED - Provider registered but not ready
• EMBEDDING_MODEL_NOT_FOUND - Model doesn't exist on provider
• PROVIDER_CONNECTION_FAILED - Network/connection to provider broken
• CONFIG_SCHEMA_MISMATCH - Configuration structure incompatible

Historically Related Issues

Related Configuration Keys

memory.embedding.provider          # The failing key
memory.embedding.model              # May need updating
memory.embedding.providerConfig     # Optional per-provider config
memory.embedding.dimensionOverride  # May conflict with model output
memory.providers.fallback            # Fallback chain configuration

Downstream Error Cascade

When this error occurs, subsequent operations fail:

openclaw memory search "query"
# → Fails: No embedding provider available

openclaw memory index
# → Fails: Cannot generate embeddings for new content

openclaw dream
# → May partially work: Uses cached embeddings