openclaw - 💡(How to fix) Fix Memory log (memory/YYYY-MM-DD.md) not generated since 4.8 upgrade — embedding API required but should not be [2 comments, 1 participants]

Bug Description

Since upgrading to OpenClaw 2026.4.8, daily memory log files (memory/YYYY-MM-DD.md) are no longer generated. The last log file was created on 2026-04-06. This is a regression — memory log generation worked fine before the upgrade.

Environment

• OpenClaw version: 2026.4.8 (previously 2026.4.5)
• OS: macOS
• Install method: npm global
• Primary model: volcengine-plan/minimax-m2.7 (configured via openclaw.json)
• Auth method: API key (no OAuth used)

Expected Behavior

Daily memory log files should be generated automatically by memory-core compaction mechanism, requiring only the LLM for summarization — not an embedding API key.

Actual Behavior

The openclaw memory status --deep command shows:

Embeddings: unavailable
Embeddings error: No API key found for provider "openai". You are authenticated with OpenAI Codex OAuth. Use openai-codex/gpt-5.4 (OAuth) or set OPENAI_API_KEY to use openai/gpt-5.4.

And:

Recall store: 0 entries · 0 promoted · 0 concept-tagged · 0 spaced
Recall path: ~/.openclaw/workspace/memory/.dreams/short-term-recall.json

No short-term recall entries are being stored. As a result:

1. openclaw memory promote --apply returns "No short-term recall candidates"
2. The daily memory log files (memory/YYYY-MM-DD.md) are not generated
3. The memory-dream cron task has nothing to process

Root Cause Analysis

The memory-core system appears to require an embedding API key for its normal operation (recall store, semantic indexing). However:

• Generating daily memory log files should only require LLM summarization, not vector embeddings
• This requirement appears to have been tightened in 4.8, breaking a function that previously worked without explicit embedding configuration
• Users with valid LLM API keys should not need a separate embedding provider just to generate daily memory logs

Configuration

openclaw.json uses volcengine-plan provider with API key auth for the primary model. No OAuth is actively used.

Questions / Considerations

1. Should daily memory log generation require an embedding provider?
2. Is this a regression from 4.8 config schema changes?
3. Is there a fallback mode for memory-core that works without embedding?

Workaround

None found. User asked to configure a separate embedding provider as a workaround.

Reported by: OpenClaw user on Feishu