Environment:

• OpenClaw image: ghcr.io/openclaw/openclaw:2026.6.1
• Runtime: Docker container, Linux
• Memory backend: builtin memory with OpenAI hybrid/vector provider
• Corpus: generated synthetic markdown memory files at 1,000 and 10,000 entry
  scales

Summary:

When indexing a large generated memory corpus as one bulk import/indexing pass,
OpenClaw hit an OpenAI 431 request_headers_too_large failure before producing
usable search results at 1,000 and 10,000 entries. Importing the same style of
corpus in 100-entry chunks and running openclaw memory index --agent main
after each chunk avoided the 431 failure and successfully indexed through
10,000 entries.

Repro shape:

1. Start an initialized OpenClaw 2026.6.1 container with builtin memory using
   the OpenAI provider.

2. Generate a memory corpus with many contact-like notes, dated events,
   preferences, and conversation summaries.

3. Place the full 1,000-entry or 10,000-entry generated memory corpus into the
   workspace memory path.

4. Run:

   openclaw memory index --agent main

Observed in the bulk path:

• 100-entry scale indexed successfully.
• 1,000 and 10,000 entry scales failed before usable search with OpenAI
  431 request_headers_too_large.

Observed in the chunked path:

• Add 100 generated entries at a time.
• Run openclaw memory index --agent main after each chunk.
• 100 chunks completed through 10,000 entries with no 431 error.
• Representative chunk duration was roughly 4.5s per 100 entries in the
  follow-up run; later full agent-path runs measured about 13 entries/s in
  this local environment.

Expected:

• Large memory imports should either batch provider calls internally or produce
  an explicit actionable error that tells operators to batch/chunk the import.
• A single large import should not form provider requests that exceed upstream
  provider header limits.

Impact:

• Tenant onboarding/history import can fail at realistic memory sizes unless
  the operator manually chunks imports.
• Managed services need predictable import throughput and retry behavior for
  customer history loads.

Notes:

• This report is based on generated synthetic data only.
• No personal state, customer data, or raw provider keys are required for the
  repro.