Commit Graph

31 Commits (de2dfd3034a7b506fa269910401bbdcbd4efc71d)

Author SHA1 Message Date
lyingbug 04adc13272
fix(retrieval): fix recall correctness, score scales, cost and agent tools (#3687)
* fix(retrieval): keep index writes and shared-KB lookups correct

Index writes
- Image OCR/caption chunks were indexed without IsEnabled, so every engine
  but Postgres stored them disabled and retrieval never returned them.
  The SQLite engine re-enables the affected rows at startup.
- FAQ edits, similar-question additions and import merges re-index under
  the old SourceID and relied on the engine overwriting it. Postgres and
  SQLite kept the stale row (ON CONFLICT DO NOTHING); ES v8 and Qdrant
  mint a new document ID and kept both. FAQ re-indexing now deletes the
  old rows first, and Postgres/SQLite BatchSave upserts by
  (source_id, source_type).
- Postgres: drop the GORM default tag on is_enabled, which replaced false
  with true on insert and indexed disabled chunks as enabled.
- SQLite: BatchSave no longer attaches vectors and FTS rows to the wrong
  rows when part of a batch already exists, and vecTables is guarded
  against concurrent map writes.
- CopyIndices: Postgres filters to the copied knowledge, pages by id and
  keeps is_enabled; SQLite copies every row of a chunk (generated
  questions included) with stable source IDs and enabled state.

Shared knowledge bases
- Chat merge: FAQ answer population and neighbor expansion look chunks up
  without the caller-tenant filter, like parent resolution (#3342).
- Neighbor expansion never cuts the retrieval hit; the length budget goes
  to the text adjacent to it.
- FAQ negative questions are applied whenever an FAQ KB is in scope and
  to chunks of shared FAQ KBs.
- Search results get image_info for shared-KB chunks.
- search_knowledge reads FAQ metadata from the search result instead of
  one tenant-scoped query per hit, so shared FAQs render their answers.

* fix(retrieval): put every retrieval score on one scale

Scores from different retrieval paths were ranked together while living on
different scales, so ordering depended on which path produced a hit.

Fusion
- RRF ranked a chunk by its position in the concatenation of every
  vector (or keyword) list, appended in goroutine completion order. The
  second list's best hit ranked behind every hit of the first, and the
  order changed between identical requests. Each list is now ranked on
  its own by score and a chunk takes its best rank.
- RRF is divided by its maximum, so hybrid scores land in [0, 1] like
  vector-only (cosine) and keyword-only (rescaled BM25) results. Raw RRF
  topped out near 0.016 and lost to any single-retriever hit it was ranked
  against across searches (document targets, model groups, FAQ calls);
  rerank's base-score term and MMR's relevance term had nothing to use.
- Keyword-only BM25 is rescaled per list; engines' BM25 is not comparable.
- FAQ iterative retrieval fuses each round like the first pass instead of
  sorting raw BM25 and cosine together, starts at twice the depth already
  searched, triggers when any single vector list is full (the old summed
  count never matched with more than one list) and replaces results per
  round.

Rerank and chat pipeline
- The FAQ boost is no longer capped at 1; capping made strong FAQs tie at
  1.0 and fall to the tie-breaker. The composite score before boosts is
  kept as composite_score.
- The FAQ direct-answer threshold is compared against that pre-boost
  score, and the FAQ that cleared it is the one marked match="exact"
  rather than the first FAQ.
- Graph hits no longer carry a fabricated retrieval score of 1.0, which
  ranked them above every scored hit when rerank did not run; the
  composite score uses the model score in its place.

Engines
- OpenSearch and Weaviate return (1 + cos) / 2; their drivers now report
  cosine similarity and map thresholds onto the engine scale, so a
  threshold means the same on every engine.
- Milvus L2 returned squared distance, ranked farthest-first downstream,
  and took the threshold as a radius; scores and radius are converted.
- The multi-store normalizer no longer maps Milvus through (s + 1) / 2,
  which put it on a different scale from every other engine and from its
  own single-store results.

Docs updated for fusion, normalization, FAQ boost and iterative recall.

* perf(retrieval): bound rerank, agent and pipeline cost; stop failing on slow or missing models

Rerank
- Rerankers report the longest document a request can carry
  (max_document_chars / max_request_chars minus the query) and the rerank
  stage fits its own passages to it, trimming appended captions, OCR and
  generated questions first. One chunk with a large screenshot's OCR used
  to exceed the vendor limit and fail the whole batch, so every candidate
  lost its score and the threshold stopped applying.
- Rerank HTTP requests default to a 60 s timeout; chat and agent contexts
  carry no deadline, so a stalled self-hosted rerank server held the turn
  open.
- OCR and caption chunks no longer repeat their own text from image_info in
  the passage.
- The chat pipeline and the agent tool rerank at most 200 candidates, best
  retrieval score first.
- A rerank model that cannot be loaded degrades to retrieval order instead
  of failing every chat turn, as the search APIs already do; without rerank
  results the merge stage keeps only RerankTopK hits instead of enriching
  every one.

Retrieval APIs
- knowledge-search match_count and rerank.top_k are capped at 200
  (unbounded values multiplied into tens of thousands of billed rerank
  passages); normalizedMatchCount clamps to the pool size so the
  over-retrieval arithmetic cannot overflow.
- hybrid-search with rerank recalls at least rerank.top_k deep.

Agent
- search_knowledge fits its rendered output to four fifths of the tool
  output budget, dropping the lowest-ranked rows and telling the model how
  many it omitted; the model's view is rebuilt from Data, which registry
  truncation never reached.
- Agent searches skip neighbor/parent/related enrichment rows, which were
  never retrieved yet tripled the rerank input.
- ParallelToolCalls is enabled. The model was always told it may call tools
  in parallel while the engine ran them one by one; only read-only tools
  overlap, anything else stays a barrier.

Chat pipeline
- Query expansion searches keywords only (its variants are local rewrites
  whose embeddings repeat the main search) and is skipped when keyword
  matching is off.
- Entity extraction, a second LLM call, is skipped for intents that need no
  KB retrieval and uses the rewritten query.
- Partial-overlap dedup tokenizes each text once instead of per pair, and
  only compares token overlap between texts within 3x of each other's size.

Storage
- Migration 000112 adds btree indexes on embeddings(knowledge_id) and
  (chunk_id); document deletes, chunk re-index and enable/tag updates
  scanned the whole table.
- Shared-KB knowledge rows are fetched in one query instead of one per ID.
- Drop GORM Debug() from the ParadeDB keyword query and the knowledge batch
  lookup, which logged SQL and user queries on every search.

Docs updated for the API limits, rerank timeout and passage fitting, agent
parallel tools and output budget, and pipeline changes.

* fix(retrieval): make agent tools and pipeline stages report what they did

Agent tools
- query_knowledge_graph now queries the graph store: entity names from the
  query are matched in each KB's graph, relations are returned (and shown to
  the model), and the entities' source chunks lead the results, with text
  search as a supplement. It used to run text search only, so a graph-only
  KB always answered "no relevant graph information".
- search_knowledge reports failures instead of "no matching chunks": when
  every search fails the tool fails with the reason; partial failures are
  listed. A failed query embedding falls back to keyword search for hybrid
  KBs, as the chat pipeline does.
- Keyword mode uses rerank to order hits, not to reject them (identifiers
  score low), and the empty-result hint no longer sends keyword mode back to
  keyword mode.
- Dedup keeps siblings under one parent, no longer merges every image chunk
  of a document through chunk_index 0, and keeps the better-scored copy.
- Search rows omit a <match> excerpt that only repeats the content; snippet
  terms are segmented for Chinese queries and skip stopwords, and FAQ
  matched questions are chosen by match strength.
- read_document query mode returns next_offset when truncated and resumes
  from offset.
- wiki_search over-fetches under document/tag scopes before filtering and
  says when every match was out of scope.
- Prompts point to read_document(id=cN, context=k) for surrounding text.
- The agent editor hides the retrieval-strategy section in agent mode,
  where search_knowledge uses the global retrieval settings.

Chat pipeline
- Intent labels are normalized; an unknown label retrieves instead of
  silently skipping the knowledge base, and web_search without web search
  enabled searches the knowledge bases.
- Image turns get a 2048-token rewrite budget, and a reply cut off by the
  cap is salvaged field by field instead of being discarded.
- History references are matched by how much of the query they cover;
  Jaccard against chunk-sized text never reached its threshold.
- Entity search resolves chunks of org-shared KBs within the turn's KB
  scope, skips chunks whose document is gone (it used to panic) and caps
  graph chunks at 30.
- Empty history is not fetched twice; fetched pages are cut on a rune
  boundary.

Retrieval
- A multi-KB search whose primary KB has no vectors embeds the query with a
  vector KB's model instead of failing on an empty model ID.
- Rerank passages include the chunk's heading breadcrumb, which is part of
  what was embedded.
- SQLite FTS searches the content column only and turns a lone Han
  character into a prefix query; TencentVectorDB ef follows TopK.

Docs updated for the agent tools and pipeline stages.

* fix(retrieval): address review findings on the retrieval fixes

- Restore NeedsRetrieval for web_search without web search enabled: the
  intent prompt "web_search" is how the user learns web search is off, and
  running KB retrieval instead bypassed it.
- query_knowledge_graph keeps the case of entity terms; Neo4j matches node
  names with case-sensitive CONTAINS, so lowercased terms never matched
  English entities.
- An empty search_knowledge or query_knowledge_graph result states partial
  failures in its text, which is all the model sees for an empty result.
- SQLite BatchSave replaces a source in one transaction, so a failed insert
  keeps the old rows instead of leaving the source unindexed.
- Reuse stringSliceValue and an exported searchutil.ContainsChinese instead
  of new copies; drop an unused named return and the FAQ obsolete-source
  delete that the full per-chunk delete now covers.

* fix(retrieval): address second review of the retrieval fixes

- query_knowledge_graph: under a document/tag scope, keep only relations
  whose entities come from an in-scope evidence chunk (the graph namespace
  is the whole KB); drop chunks whose document no longer exists; report a
  graph or text search failure even when the other source returned hits.
- search_knowledge: record the hybrid -> keyword fallback taken when the
  query embedding fails in mode_fallbacks, so an empty result is not read
  as "no semantic match".
- query understanding salvage: take a cut-off value only for the image
  description; a truncated rewrite_query or intent is dropped instead of
  replacing the user's query.
- Graph hits (no retrieval score) are kept past the no-rerank RerankTopK
  cut and the rerank candidate cap instead of always falling below them.
- FAQ boost is capped at 1 again so scores stay on the [0, 1] scale; FAQs
  tied at the cap are ordered by their pre-boost score.
- Negative-question filtering loads only candidates of FAQ knowledge bases.
- read_document query continuation starts one chunk before offset so the
  first match keeps its preceding context at a page boundary.
- graphSearchTerms and searchQueryTokens share one queryTerms splitter.
2026-09-25 10:48:18 +08:00
lyingbug e179f44445
refactor(models): separate provider definitions, model catalogs, and runtime configuration (#3589)
* refactor(models): separate providers, catalog and runtime configuration

* refactor(models): isolate provider definitions and runtime catalogs

* docs: exclude refactoring analysis notes from the pull request
2026-09-23 14:20:08 +08:00
lyingbug 4640a1ea0c
test(rerank): pin every vendor's request, and correct three facts the table surfaced (#3513) 2026-09-21 20:01:09 +08:00
lyingbug 41dde3b05f
refactor(models): delete the internal/models/provider façade (#3506) 2026-09-21 14:41:48 +08:00
lyingbug d6c7665cb5
refactor(rerank): move the wire formats into protocol packages, and correct the vendor facts (#3488)
把 8 个手写 rerank client 收敛成 3 个协议包(internal/models/api/{cohererank,dashscoperank,nimrerank}),每个带一份从厂商文档逐字粘来的 golden fixture。LKEAP 和火山保留官方 SDK,但只实现 api.Reranker。rerank.NewReranker 改成 catalog.Resolve + 按协议分发,函数体里没有厂商名。生产代码净 -147 行。

三件原来每个 client 各管各、现在由共用层统一的事:分批(按厂商声明的上限,智谱 128 / NIM 512 等此前无人执行)、分数标度(NIM 与 GPUStack 返回 logit,其余为 0~1 概率,统一换算后再与 RerankThreshold 比较)、NIM 的 truncate(默认 NONE 会让超长 passage 整请求失败,此前从未发送)。

对照官方文档订正四处厂商事实:GPUStack 分数标度、千帆三项上限、阿里 500 条上限、qwen3-rerank 是第二种协议(隐藏并在构造期拒绝)。Novita / OpenRouter / LiteLLM 三家官方有 rerank 但未开类型,本次开放,零协议代码。

运行时行为变更:NVIDIA 起发 truncate: END;GPUStack 分数按 logit 换算(可按行覆盖);truncate_prompt_tokens 收窄为厂商声明制;qwen3-rerank 存量行从静默打错变为构造期报错。详见 PR 描述。

关联 #3486。
2026-09-21 10:36:26 +08:00
lyingbug c40a9dd194
refactor(models): rebuild vendor integration on protocol + catalog + vendor layers (#3470)
把集中式的 OpenAI 方言客户端重建为三层:协议层 internal/models/api/<protocol>(OpenAI Chat Completions / OpenAI Responses / Anthropic Messages / Google GenAI),目录层 internal/models/catalog,厂商层 internal/models/vendors/<id>(27 家,go:embed)。27 家厂商逐条对照官方文档核验,厂商事实以数据声明,协议层不出现厂商名。

既有行的运行时行为会变的四处:
1. extra_config.api 变为保留的协议选择器,非法取值被拒绝;
2. Azure 未填 api_version 的行改走 /openai/v1 GA 数据面;
3. api.openai.com 的一方流量改走 Responses 协议(中转/网关不受影响);
4. 7 家厂商的输出上限字段按文档纠正。

不需要数据库迁移。详见 PR 描述与 website-docs/03-features/06-models.md。
2026-09-20 21:28:23 +08:00
wizardchen ffcb10f953 feat(memory): add cross-session long-term memory
Persist extracted and explicit memories across sessions with topic
resolution, semantic recall, document affinity, consolidation, workspace
settings, and inbox/interests UI.
2026-08-13 18:00:02 +08:00
xiaowang 31a2358bbf fix(rerank): normalize NVIDIA logit scores 2026-08-03 15:07:15 +08:00
xiaowang fe65633f38 fix(rerank): batch LKEAP requests within API limits 2026-08-03 14:59:16 +08:00
wizardchen 146e341748 refactor(http): share SSRF-safe transport across model clients
Extract NewSSRFSafeTransport and NewSSRFSafeHTTPClientWithTransport so
embedding and rerank can reuse one connection pool without allocating a
throwaway transport per client. SSRF dial and redirect policies are
unchanged; each client still keeps its own timeout.
2026-07-23 12:00:05 +08:00
Martín Alcalá Rubí 83da0903b6 fix(rerank): stop sending vLLM truncate_prompt_tokens by default
The generic OpenAI-compatible reranker hardcoded `truncate_prompt_tokens:
511` on every /rerank request. Providers that honor this vLLM-specific
field (e.g. SiliconFlow serving Qwen3-Reranker) keep only the last 511
tokens of the templated rerank prompt, cutting the instruction and query
off any document longer than ~511 tokens. Every relevance score then
collapses to near zero (~0.0067 instead of ~0.998), falls below the score
threshold, and knowledge-base search returns empty results. A direct curl
without the field returns the correct scores.

WeKnora already removed this exact hardcoded parameter from the embedding
path in #960; the rerank path was never fixed.

- Add `omitempty` to `truncate_prompt_tokens` and `additional_data` so
  empty optional fields are not sent.
- Make truncation opt-in via `extra_config.truncate_prompt_tokens`
  (validated > 0 at construction), preserving an escape hatch for
  self-hosted vLLM deployments with small-context rerankers.
- Add regression tests reproducing the collapsed-score symptom and the
  opt-in path.

Fixes #2143
2026-07-21 12:12:56 +08:00
wizardchen b1b42abe23 fix(rerank): batch Volcengine rerank requests over API document limit
Split oversized candidate sets into concurrent batches instead of failing when retrieval exceeds the VikingDB rerank per-request cap, and align ko/ru locale hints with zh-CN.
2026-07-20 20:07:21 +08:00
KKtwo 3011f132e9 feat: support Volcengine rerank 2026-07-20 19:58:48 +08:00
wizardchen 1661827d6e fix(security): harden model client redirects 2026-07-02 19:15:46 +08:00
wizardchen de525a4f02 feat(observability): enhance Langfuse retrieve and rerank tracing
Add shared retrieval observation helpers, expand rerank generation output
with previews and score stats, and enrich retrieve/rerank pipeline spans.
2026-06-16 01:09:23 +08:00
wizardchen dda16436d0 feat(lkeap): integrate Tencent Cloud LKEAP Rerank support
- Added support for Tencent Cloud LKEAP Rerank, including new configuration options for SecretId and SecretKey.
- Updated frontend components to handle LKEAP-specific fields and enhance user experience with appropriate labels and placeholders.
- Enhanced backend logic to manage LKEAP credentials and integrate with the new rerank API.
- Improved localization for LKEAP-related terms across multiple languages, ensuring clarity and consistency in user-facing messages.
- Introduced tests for LKEAP Reranker functionality to ensure reliability and correctness in handling requests.
2026-06-04 17:22:18 +08:00
wizardchen 492e92580b feat(observability): integrate Langfuse for LLM token tracking and tracing
Closes #620 #497. Add opt-in Langfuse observability covering all five
model types (chat, embedding, rerank, VLM, ASR) with HTTP-request-scoped
traces and Docker Compose support (both cloud and self-hosted).

Core package internal/tracing/langfuse:
- HTTP client with batched async ingestion (non-blocking in request path)
- Sampling, environment / release tagging, and graceful fallback when
  LANGFUSE_* env vars are absent (wrappers become no-ops)
- Gin middleware opens one trace per traced request and finishes it after
  the handler chain returns, attaching method / path / user / session
- Trace context is stored under a typed key exported from internal/types
  so logger.CloneContext can preserve it across handler / goroutine
  boundaries (otherwise each LLM call auto-created an orphan trace,
  fragmenting one request into many)

Per-model generation wrappers (opt-in via NewChat/NewEmbedder/...):
- chat: captures prompt, streaming output, token usage + TTFT
- embedding: approximates tokens when the provider omits usage
- rerank: previews query/docs, summarizes results to keep payload small
- vlm: records image count and total bytes, never uploads raw pixels
- asr: records file size and audio duration, never uploads audio bytes

Async title generation (GenerateTitleAsync) now forwards the trace key
into the goroutine so title calls appear under the parent chat trace.

Docker Compose:
- LANGFUSE_* env passthrough on the `app` service for cloud deployments
- Optional `langfuse` profile spins up a self-hosted Langfuse stack that
  reuses WeKnora's existing PostgreSQL (separate database via an idempotent
  init container that fixes ICU collation drift) and Redis (separate DB
  number), adding only ClickHouse, MinIO, web and worker containers
- web/worker entrypoints URL-encode DB_PASSWORD / REDIS_PASSWORD at start
  to avoid Prisma P1013 when passwords contain @ / # / etc.

Docs: docs/Langfuse集成.md covers cloud vs self-hosted, per-model usage
strategy, code map, and resource footprint.
2026-04-24 10:29:19 +08:00
wizardchen fd3e2992b1 feat(models): support custom HTTP headers across all remote model calls
Introduce per-model `custom_headers` config (similar to OpenAI Python SDK's
`extra_headers`) so users can inject gateway auth tokens, trace IDs, etc.
into every outbound API request for chat / embedding / rerank / VLLM / ASR
models. Reserved headers like Authorization / Content-Type are always
preserved to avoid breaking auth or signing flows.

Along the way, unify production and "test connection" paths onto a single
`ConfigFromModel(*types.Model, appID, appSecret)` constructor in each
model package. Both `service.modelService.GetXxxModel` and the four
`handler.initialization.Check*Model` / `TestEmbeddingModel` endpoints now
go through the same field mapping, so new parameters only need to be
added in one place.

Backend:
- Add `CustomHeaders map[string]string` to `types.ModelParameters` with
  JSON/YAML `omitempty` tags for forward compatibility.
- New `internal/utils/extraheaders.go` providing `ApplyCustomHeaders`
  (for hand-rolled HTTP paths) and `WrapHTTPClientWithHeaders` /
  `CustomHeadersRoundTripper` (for SDK paths like go-openai). Reserved
  headers (Authorization, api-key, Content-Type, Accept, Host,
  Content-Length, User-Agent) are filtered out.
- Inject headers in every remote model path: chat (SDK + raw HTTP),
  embedding (openai/aliyun/jina/volcengine/nvidia/azure_openai), rerank
  (remote_api/aliyun/jina/nvidia/zhipu — via shared `customHeaderSetter`
  interface), VLM and ASR (via http.Client wrapping).
- Add `ConfigFromModel` to each of chat/embedding/rerank/vlm/asr,
  consolidating field mapping (ExtraConfig, CustomHeaders, InterfaceType
  defaulting, WeKnoraCloud credentials, etc.) and cover it with
  dedicated unit tests per package.
- Refactor `service/model.go`: replace ~100 lines of hand-written Config
  literals with one-line `ConfigFromModel` calls; drop the now-unused
  `stringMapToAnyMap` helper.
- Refactor `handler/initialization.go` test-connection endpoints: merge
  four ad-hoc request structs into one `ModelTestRequest`, extract
  `buildTestModel` and `resolveTenantWeKnoraCloudCreds` helpers, and
  route all four endpoints through `ConfigFromModel` + `NewXxx` so the
  test path is now behaviorally identical to production.

Frontend:
- Add `custom_headers` to the `ModelConfig` API type.
- `ModelEditorDialog.vue`: new Key-Value header editor (add/remove rows),
  auto-converts map <-> array when loading/saving, and passes the
  current header set to the `Test Connection` button so the preview
  reflects exactly what production will send.
- `ModelSettings.vue`: serialize the header array back into a map for
  the backend (drops empty rows).
- i18n: add custom-header labels / descriptions / placeholders to
  zh-CN, en-US, ru-RU, ko-KR.

Tests:
- `internal/utils/extraheaders_test.go` covers reserved-header filtering
  and round-tripper wrapping (including nil-client fallback).
- New `config_from_model_test.go` in chat / embedding / rerank / vlm /
  asr verifies all fields (CustomHeaders, ExtraConfig, InterfaceType
  defaulting, AppID/AppSecret) are propagated end-to-end.
2026-04-23 23:36:40 +08:00
wizardchen 53e8e17df4 feat(logger): implement LLM debug logging functionality
- Added a new logger for LLM calls, enabling detailed logging of model interactions, including request and response data.
- Introduced configuration options for enabling and specifying the log directory for LLM debug logs.
- Implemented cleanup for old log files to manage disk space effectively.
- Wrapped existing chat implementations to log calls when debug logging is enabled, enhancing traceability for model interactions.
2026-04-16 23:15:02 +08:00
begoniezhao 2c0c1b8bab feat: integrate ChatBot provider and docreader 2026-04-13 17:47:24 +08:00
wizardchen 5fe1ccd4f9 feat: enhance multimodal support and image handling
- Updated configuration to improve handling of multimodal queries, including clearer guidelines for when to use [NO_SEARCH] in responses.
- Enhanced the user interface to include dedicated sections for multimodal configuration, allowing users to manage image uploads and related settings more effectively.
- Added localization support for new multimodal features in both English and Chinese, improving accessibility for a broader audience.
- Implemented backend changes to support image analysis and extraction of OCR text, enriching the interaction experience for users.

These enhancements significantly improve the application's capability to process and respond to multimodal queries, fostering a more interactive and user-friendly environment.
2026-03-16 02:26:47 +08:00
Manx98 5afc81d00b feat: support NVIDIA model API 2026-03-13 18:14:08 +08:00
nullkey e9a2a7b3e5 feat: 新增对多家平台的模型适配 2025-12-29 12:18:42 +08:00
wizardchen eb8bfddfd1 refactor: Update function signatures across multiple files to improve readability by adding context parameters and enhancing code structure 2025-11-28 16:17:52 +08:00
wizardchen 135e33eca4 refactor: Clean up code by removing unnecessary blank lines and improving comments for better readability and maintainability 2025-11-28 16:10:37 +08:00
wizardchen bf35a2861c refactor: Sanitize log inputs to prevent injection attacks 2025-11-25 22:19:41 +08:00
wizardchen 2b6cbee1b6 feat: add aliyun rerank 2025-09-10 01:22:25 +08:00
wizardchen 4214e6782b feat: add aliyun rerank model 2025-09-10 01:22:25 +08:00
wizardchen 2e029731ed feat: Compatibility with rerank API output fields 2025-09-08 11:20:40 +08:00
Liwx1014 19c5e242bf 兼容解析重排序服务返回score字段 2025-08-28 16:27:59 +08:00
wizardchen 56eb2bce33 init commit 2025-08-05 15:08:07 +08:00