openclaw - ✅(Solved) Fix [Bug]: Memory index fails with "embeddings max length per batch size is 2000" — no chunking before embedding [2 pull requests, 1 comments, 2 participants]

openclaw2026-05-10 10:47:15

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

openclaw/openclaw#80226•Fetched 2026-05-11 03:17:21

View on GitHub

Comments

Participants

Timeline

Reactions

Author

gaodaabao

Participants

clawsweeper[bot]

gaodaabao

Timeline (top)

cross-referenced ×3labeled ×2commented ×1referenced ×1

Error Message

openclaw memory index --force fails with embedding API error. OpenClaw sends entire files to the embedding API without pre-chunking. When combined batch exceeds model's token limit, entire index fails — zero chunks stored. Files should be pre-chunked to <=model's max tokens before embedding. Single-file failures should warn and skip, not crash the entire index. Memory index failed (main): openai embeddings failed: 400 {"error":{"code":"invalid_argument","message":"embeddings max length per batch size is 2000","type":"invalid_request_error"}} Memory index failed (main): openai embeddings failed: 400 {"error":{"code":"invalid_argument","message":"embeddings max length per batch size is 2000","type":"invalid_request_error"}}

Root Cause

Fix Action

Fixed

Fixed by PR: [codex] Fix memory embedding batch limit handling (https://github.com/openclaw/openclaw/pull/80241)
Fixed by PR: fix(memory): handle per-batch item limits and add retry for 2000-item errors (https://github.com/openclaw/openclaw/pull/80423)

PR fix notes

PR #80241: [codex] Fix memory embedding batch limit handling

Repository: openclaw/openclaw
Author: NianJiuZst
State: open | merged: False
Link: https://github.com/openclaw/openclaw/pull/80241

Description (problem / solution / changelog)

Summary

Split inline memory embedding batches again when a provider rejects the combined batch as too large.
Skip only irreducible input-size embedding failures per file, while preserving provider-level failures as command failures.
Document the OpenAI-compatible low-limit behavior and add regression coverage for issue #80226.

Fixes #80226.

Verification

pnpm test packages/memory-host-sdk/src/host/embedding-chunk-limits.test.ts extensions/memory-core/src/memory/manager-embedding-policy.test.ts extensions/memory-core/src/memory/index.test.ts extensions/openai/embedding-provider.test.ts -- --reporter=verbose
pnpm test extensions/memory-core/src/memory/manager-sync-ops.archive-delta-bypass.test.ts extensions/memory-core/src/memory/manager-sync-yield.test.ts -- --reporter=verbose
pnpm tsgo:extensions
pnpm tsgo:test:extensions
pnpm tsgo:test:packages
git diff --check
pnpm docs:list

Changed files

CHANGELOG.md (modified, +1/-0)
docs/reference/memory-config.md (modified, +2/-0)
extensions/memory-core/src/memory/index.test.ts (modified, +72/-0)
extensions/memory-core/src/memory/manager-embedding-ops.ts (modified, +75/-13)
extensions/memory-core/src/memory/manager-embedding-policy.test.ts (modified, +13/-0)
extensions/memory-core/src/memory/manager-embedding-policy.ts (modified, +6/-0)
extensions/memory-core/src/memory/manager-sync-ops.archive-delta-bypass.test.ts (modified, +7/-1)
extensions/memory-core/src/memory/manager-sync-ops.ts (modified, +30/-2)
extensions/memory-core/src/memory/manager-sync-yield.test.ts (modified, +7/-1)
extensions/openai/embedding-provider.test.ts (modified, +20/-0)
packages/memory-host-sdk/src/engine-embeddings.ts (modified, +1/-0)
src/plugin-sdk/memory-core-host-engine-embeddings.ts (modified, +1/-0)

PR #80423: fix(memory): handle per-batch item limits and add retry for 2000-item errors

Repository: openclaw/openclaw
Author: dlrik
State: closed | merged: False
Link: https://github.com/openclaw/openclaw/pull/80423

Description (problem / solution / changelog)

Summary

Qianfan bge-large-zh (and some other providers) enforce a max 2000 items per embedding batch. Previously, when buildMemoryEmbeddingBatches produced a batch exceeding this limit, the entire index failed with:

embeddings max length per batch size is 2000

Two-part fix:

Subdivide token-based batches by item count: Added EMBEDDING_BATCH_MAX_ITEMS = 2000 and subdivide token-based batches into sub-batches of at most 2000 items before calling the provider.
Retry on item-count limit errors: Added max length per batch and batch size is N to isRetryableMemoryEmbeddingError so they are retried with automatic subdivision.

Test plan

Verify memory index with Qianfan/bge-large-zh no longer fails with 2000-item batches
Verify existing token-based batching still works for providers without item limits
Verify error is still retryable and crashes the index when subdivision also fails

Fixes #80226

Changed files

extensions/memory-core/src/memory/manager-embedding-ops.ts (modified, +25/-16)
extensions/memory-core/src/memory/manager-embedding-policy.ts (modified, +1/-1)
src/agents/pi-embedded-runner/run.ts (modified, +2/-1)
src/agents/pi-embedded-runner/run/payloads.ts (modified, +11/-0)
src/gateway/open-responses.schema.ts (modified, +15/-0)
src/gateway/protocol/schema/logs-chat.ts (modified, +2/-0)
src/gateway/server-methods/chat-transcript-inject.ts (modified, +3/-1)
src/gateway/server-methods/chat.ts (modified, +4/-0)

Code Example

Memory index failed (main): openai embeddings failed: 400 {"error":{"code":"invalid_argument","message":"embeddings max length per batch size is 2000","type":"invalid_request_error"}}

RAW_BUFFERClick to expand / collapse

Bug type

Regression (worked before, now fails)

Beta release blocker

Summary

Steps to reproduce

Configure agents.defaults.memorySearch with Qianfan bge-large-zh
Have any memory files totaling >2,000 tokens in combined batch
Run openclaw memory index --force
Result: entire index fails, 0 chunks indexed

Expected behavior

Files should be pre-chunked to <=model's max tokens before embedding. Single-file failures should warn and skip, not crash the entire index.

Actual behavior

Memory index failed (main): openai embeddings failed: 400 {"error":{"code":"invalid_argument","message":"embeddings max length per batch size is 2000","type":"invalid_request_error"}}

OpenClaw version

2026.5.6

Operating system

centos8

Install method

npm

Model

Baidu Qianfan (OpenAI-compatible endpoint) /bge-large-zh

Provider / routing chain

openclaw Baidu Qianfan /bge-large-zh

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Memory index failed (main): openai embeddings failed: 400 {"error":{"code":"invalid_argument","message":"embeddings max length per batch size is 2000","type":"invalid_request_error"}}

Impact and severity

No response

Additional information

No response

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

FAQ

Expected behavior

Files should be pre-chunked to <=model's max tokens before embedding. Single-file failures should warn and skip, not crash the entire index.

#api #training loop #device allocation #model download #tokenizer error

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

openclaw - ✅(Solved) Fix [Bug]: Memory index fails with "embeddings max length per batch size is 2000" — no chunking before embedding [2 pull requests, 1 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Error Message

Root Cause

Fix Action

Fixed

PR fix notes

PR #80241: [codex] Fix memory embedding batch limit handling

Description (problem / solution / changelog)

Summary

Verification

Changed files

PR #80423: fix(memory): handle per-batch item limits and add retry for 2000-item errors

Description (problem / solution / changelog)

Summary

Test plan

Changed files

Code Example

Bug type

Beta release blocker

Summary

Steps to reproduce

Expected behavior

Actual behavior

OpenClaw version

Operating system

Install method

Model

Provider / routing chain

Additional provider/model setup details

Logs, screenshots, and evidence

Impact and severity

Additional information

FAQ

Expected behavior

Still need to ship something?

RELATED_DISCOVERY

TRENDING