An embedding row with spec.input image and no image_format or image_field
was treated as text-only on every vendor except generic and Jina. Default
that case to the vLLM messages format. An explicit object format without a
field stays refused.
Fixes#4343
BatchEmbedWithPool fans a batch out over the ants pool, so with
BATCH_EMBED_SIZE=1 every text becomes its own sub-batch and several workers run
at once. Each worker starts by testing firstErr to see whether an earlier
sub-batch already failed, but that read was unsynchronised while the writes are
under mu. The race detector reports it between the guard and the failure branch.
The guard exists to stop billing a provider for work whose result will be
discarded. A worker that misses the write issues the request anyway, so the race
defeats the protection it was added for.
Batch independent Qwen3-VL embedding conversations through an explicit compatibility setting. Preserve singleton requests and serial overrides, validate response ordering, and add wire tests and opt-in live numerical parity checks.
Live verification confirms valid batched responses but does not meet strict numerical parity bounds; document the deployment-side reproducibility limitation.
* fix(models): align configuration defaults and multimodal workflows
* feat(image): enhance image retrieval diagnostics and filtering
- Implemented `retrieveObservedImagePool` to observe engine calls and record image recall decisions without altering results.
- Added detailed diagnostics in `summarizeImagePool` and `summarizeImageCandidateCut` to track image filtering decisions and scores.
- Introduced tests for image pool diagnostics to ensure accurate filtering and decision recording.
- Updated existing functions to utilize new retrieval and summarization methods for improved observability and accuracy in image handling.
* feat(embeddings): enhance image handling with new formats and message support
- Introduced `EmbeddingImageFormat` to manage image input formats, allowing for both object and messages formats.
- Updated `EmbeddingsSettings` to include new fields for `ImageFormat`, `ImagePrompt`, and `TextAsMessages` to support multimodal embedding workflows.
- Enhanced `Client` methods to build request bodies for images in both formats, ensuring compatibility with vLLM's requirements.
- Added tests to validate the new image message handling and ensure correct request formatting for different scenarios.
- Updated documentation to reflect changes in image input handling and configuration options.
* feat(similarity): add embedding similarity mode for enhanced document scoring
- Introduced a new `similarity` option in `ModelDebugOptions` to score documents and an optional image against a query.
- Updated `ModelDebugDrawer.vue` to support a new input mode for similarity, allowing users to select similarity as an input type.
- Enhanced the UI to display similarity results, including ranked candidates based on cosine similarity.
- Added tests to validate the functionality of the new similarity mode in embedding models.
- Updated localization files to include new strings related to similarity mode in multiple languages.
* feat(embeddings): add Qwen3-VL-Embedding support for multimodal input
- Introduced Qwen3-VL-Embedding model with support for both text and image inputs.
- Updated model configuration files to include compatibility settings for the new embedding type.
- Enhanced tests to validate the functionality of Qwen3-VL-Embedding, ensuring correct handling of inputs and outputs.
- Updated documentation to reflect the new model's capabilities and usage instructions.
* test(models): wrap embedding assertions to satisfy lint
* test(opensearch): isolate HTTP transports between parallel tests
* feat(models): expose embedding input modalities
Whether an embedding model maps images into the same space as text is
what will decide if a knowledge base can store image vectors. The
catalog already had the field, but most multimodal embedding models
were missing it and embedding rows never reported it.
- Catalog: declare input modalities for the Doubao and Tongyi vision
embeddings, the Aliyun multimodal / qwen-vl embedding families and
jina-embeddings-v4, as listed in each vendor's documentation. The
migration baseline changes for exactly these nine entries.
- Model responses resolve capabilities for embedding rows too, so
capabilities.input reflects the catalog merged with spec.input.
- Model editor: an "Image input" switch for embedding models. It follows
the catalog and writes spec.input only when the row differs from it;
picking another catalog model or switching vendor drops the
declaration. Model cards show the image badge for such models.
* feat(embedding): embed images with multimodal embedding models
Adds the image side of an embedder, for models that map images into the
same space as their text vectors so a text query can find an image.
- api: EmbedImage (bytes + MIME type, sent inline as base64 because the
object store is usually unreachable from the vendor) and an
ImageEmbedder implemented by all four protocols, each in the image
shape its reference documents: Ark image_url, DashScope contents[].image,
Gemini inlineData parts, and on the OpenAI shape an object keyed by the
vendor's image_field (Jina: "image"). A plain OpenAI-compatible
endpoint names no image field and takes no images.
- Settings: image_field, max_image_batch_size (default one per request),
max_image_bytes and image_mime_types, filled from vendor docs for the
DashScope multimodal models, gemini-embedding-2 and Jina.
- embedding: an optional ImageEmbedder interface reached through
AsImageEmbedder, so existing Embedder implementations and fakes are
untouched. An embedder takes images only when the catalog (or the
row's spec.input) lists image AND the endpoint can carry one. Images
are checked against the documented limits before any request, then
split at the documented images per request. The concurrency, debug
and Langfuse decorators forward it; logs and traces carry the MIME
type and size, never the bytes.
The migration baseline changes only for the entries whose resolved
embedding settings now carry image limits.
* feat(knowledge): index images by their own vector and recall them
When the knowledge base's embedding model embeds images, each image now
gets a vector of its own next to its caption and OCR chunks, so a query
can find an image by what it shows even where the caption left it out.
Ingestion
- After the caption/OCR chunks are indexed, the multimodal task embeds
the image (embedding.PrepareImage converts and shrinks it to the
vendor's documented limits) and stores it as a new image_vector chunk:
Content is the caption (OCR text when there is none) for reranking and
the answer context; its index row carries the image's vector with
SourceType ImageSourceType. Scanned PDF pages are skipped, their OCR
text serves better. A failure here is logged and never fails the task.
- The image vector gets its own chunk rather than a second index row on
the caption chunk: OpenSearch and Weaviate key documents by ChunkID,
so a second row would overwrite the caption's text vector, and copy,
move and enable/disable all follow chunk IDs.
- CompositeRetrieveEngine.BatchIndexVectors writes precomputed vectors
through each engine's normal BatchIndex, keeping row order (the
composite's de-duplication does not).
- Paths that rebuild vectors from text leave image_vector alone:
syncChunkIndex only syncs its enabled flag, updateChunkVector skips
it, and a caption edit updates its Content only.
- The multimodal task's caption/OCR index rows now set IsEnabled; only
backends with a column default were storing them as enabled.
Retrieval
- A text query scores lower against an image than against text, so when
the model embeds images the document vector search is widened by half
(capped at 500) and run at min(VectorThreshold, 0.1); each store
group's hits are then held to their own threshold before score
normalization: text at VectorThreshold, images at min(VectorThreshold,
0.1). Keyword hits on image rows are dropped, their Content being the
already-indexed caption.
- image_vector hits are searchable and merge like caption hits: parent
context is attached, and the caption is not repeated from ImageInfo.
Adds golang.org/x/image for scaling and WebP/BMP decoding.
* fix(models): report no thinking levels for embedding, rerank and ASR models
Embedding rows now carry capabilities, and the end-to-end run showed them
listing off/auto/minimal/low/medium/high: these references resolve no chat
protocol, and an empty level map reads as "every level supported". A model
without a chat protocol does not think, so its capabilities say so.
* fix(compaction): never read a cancelled turn's stream as a summary
streamSummary selects on the turn's context and the provider stream at
once. When the turn is already cancelled both are ready (providers close
the stream once their context ends), and select picks at random, so a
stopped turn could come back as a completed, partial summary with a nil
error. TestSummarizationIsNotRetriedOnceTheTurnIsCancelled caught it
intermittently in CI and reproduces under -race with several -cpu values.
Check the context before each receive and again when the stream closes.
* feat(rerank): score images with multimodal rerank models
Rerank models that read images can now be sent an image document and
score it against a text query on the same scale as text, which is what a
recall made by an image's own vector needs: a text reranker only sees the
caption, and rejects an image for what the caption left out.
- api: ImageReranker, implemented by the Cohere and NIM shapes. Image
documents go in requests of their own, so text requests keep their
exact shape. NIM sends image-only passages without text, as base64 data
URLs (its VLM reranking reference requires both). The Cohere shape sends
{"<image_field>": data URI} documents and asks for no echo of them.
- The image settings embeddings already had (image_field,
max_image_batch_size, max_image_bytes, image_mime_types) move into a
shared ImageInput embedded in both settings; resolved embedding settings
serialize exactly as before.
- Catalog: jina-reranker-m0 and nvidia/llama-nemotron-rerank-vl-1b-v2 take
text and image. Jina names "image" documents; its base64 form is not
shown in any reranker example, so the data URI its embeddings reference
documents is used and marked unverified. NVIDIA lists its six formats
and 512 passages per request.
- rerank: an optional ImageReranker reached through AsImageReranker, gated
on the model's catalog input and the endpoint; images are checked against
the documented limits before any request, scores are normalized like
text, and the debug and Langfuse decorators forward it without logging
image bytes.
- Image preparation (format conversion, shrink to fit) moves from the
embedding package to internal/models/imageprep, since rerank needs it
too.
* feat(retrieval): keep a vector hit's own similarity through fusion
Hybrid fusion replaces a hit's Score with a rank-based RRF score. That is
right for ordering, but it erases the evidence reranking needs to judge an
image found by its own vector: how close the vector actually was. Record
the engine's similarity on each vector hit (IndexWithScore.VectorScore)
before normalization and fusion rewrite Score, and carry it onto the
search result (SearchResult.VectorScore, not serialized).
* feat(reranking): judge image hits by their image, not only their caption
An image found by its own vector was reranked on its caption and OCR text,
so a text reranker rejected it for exactly what the caption left out.
- With a reranker that reads images and a LoadImage callback, image_vector
hits are scored by their image (prepared to the model's limits) on the
same scale as text, and pass or fail the same threshold. A hit whose
image cannot be read, prepared or scored falls back to its text.
- With a text reranker, its verdict stands for images that are mostly
text (OCR of 200 characters or more). A pictorial hit it rejects is kept
when its vector similarity (VectorScore) reaches Options.ImageKeepScore:
at most two, strongest vector first, appended after the ranked results
and marked kept_by=image_vector. This also returns an image when no text
passes at all.
- Diagnostics report images_scored and images_kept.
* feat(rerank): read image hits' images in every rerank path
The chat pipeline, the agent search_knowledge tool and the hybrid-search
API now pass the rerank stage a way to read an image hit's image and an
image keep score, so a multimodal reranker scores images by what they show
and a text reranker can no longer drop a pictorial image its caption
undersells.
- KnowledgeBaseService.ReadChunkImage reads the first image of a result's
image_info from the storage of its knowledge base (the knowledge base's
tenant, so shared knowledge bases read from their owner's storage),
capped at 32 MB. Plain http(s) images go through the SSRF-safe
downloader.
- The backend resolution the image multimodal task used moves into a
shared knowledgeImageReader, so ingestion and reranking find an image in
the same place.
- reranking.ImageKeepScoreFor: an image is kept over a text reranker only
when its vector was at least as close as a text hit had to be, and never
below 0.25 (empirical).
* feat(chat): show a vision model the images its answer rests on
An image recalled by its own vector reached the answer as its caption only,
so a vision model could not answer from what the caption left out.
- Knowledge QA: INTO_CHAT_MESSAGE reads, for a model that supports vision,
the images of the contexts that rest on an image vector match (at most
three distinct ones, prepared to what every vision API takes: PNG, JPEG,
WebP or GIF within 5 MB), attaches them after the user's own images, and
names the contexts they belong to.
- Agent: search_knowledge attaches the same images when the agent's model
supports vision (the agent describes tool images to models that cannot
see them, and these already carry captions). Stored history drops them;
they are read again from storage.
- An image hit and its caption hit read the same once merged, and
de-duplication keeps either. The kept copy now inherits the image
(image_vector_match), in the pipeline's exact and partial-overlap
de-duplication and in the agent tool's.
- Results the rerank stage kept outside its top-k (kept_by) no longer fall
to the later top-k cuts: the pipeline's FilterTopK and the agent tool's
MMR and limit let them ride along after the ranked results.
* docs(website): describe how image hits are reranked and shown to vision models
* Fix image evidence preservation and model source mapping
* feat(models): expose embedding input modalities
Whether an embedding model maps images into the same space as text is
what will decide if a knowledge base can store image vectors. The
catalog already had the field, but most multimodal embedding models
were missing it and embedding rows never reported it.
- Catalog: declare input modalities for the Doubao and Tongyi vision
embeddings, the Aliyun multimodal / qwen-vl embedding families and
jina-embeddings-v4, as listed in each vendor's documentation. The
migration baseline changes for exactly these nine entries.
- Model responses resolve capabilities for embedding rows too, so
capabilities.input reflects the catalog merged with spec.input.
- Model editor: an "Image input" switch for embedding models. It follows
the catalog and writes spec.input only when the row differs from it;
picking another catalog model or switching vendor drops the
declaration. Model cards show the image badge for such models.
* feat(embedding): embed images with multimodal embedding models
Adds the image side of an embedder, for models that map images into the
same space as their text vectors so a text query can find an image.
- api: EmbedImage (bytes + MIME type, sent inline as base64 because the
object store is usually unreachable from the vendor) and an
ImageEmbedder implemented by all four protocols, each in the image
shape its reference documents: Ark image_url, DashScope contents[].image,
Gemini inlineData parts, and on the OpenAI shape an object keyed by the
vendor's image_field (Jina: "image"). A plain OpenAI-compatible
endpoint names no image field and takes no images.
- Settings: image_field, max_image_batch_size (default one per request),
max_image_bytes and image_mime_types, filled from vendor docs for the
DashScope multimodal models, gemini-embedding-2 and Jina.
- embedding: an optional ImageEmbedder interface reached through
AsImageEmbedder, so existing Embedder implementations and fakes are
untouched. An embedder takes images only when the catalog (or the
row's spec.input) lists image AND the endpoint can carry one. Images
are checked against the documented limits before any request, then
split at the documented images per request. The concurrency, debug
and Langfuse decorators forward it; logs and traces carry the MIME
type and size, never the bytes.
The migration baseline changes only for the entries whose resolved
embedding settings now carry image limits.
* feat(knowledge): index images by their own vector and recall them
When the knowledge base's embedding model embeds images, each image now
gets a vector of its own next to its caption and OCR chunks, so a query
can find an image by what it shows even where the caption left it out.
Ingestion
- After the caption/OCR chunks are indexed, the multimodal task embeds
the image (embedding.PrepareImage converts and shrinks it to the
vendor's documented limits) and stores it as a new image_vector chunk:
Content is the caption (OCR text when there is none) for reranking and
the answer context; its index row carries the image's vector with
SourceType ImageSourceType. Scanned PDF pages are skipped, their OCR
text serves better. A failure here is logged and never fails the task.
- The image vector gets its own chunk rather than a second index row on
the caption chunk: OpenSearch and Weaviate key documents by ChunkID,
so a second row would overwrite the caption's text vector, and copy,
move and enable/disable all follow chunk IDs.
- CompositeRetrieveEngine.BatchIndexVectors writes precomputed vectors
through each engine's normal BatchIndex, keeping row order (the
composite's de-duplication does not).
- Paths that rebuild vectors from text leave image_vector alone:
syncChunkIndex only syncs its enabled flag, updateChunkVector skips
it, and a caption edit updates its Content only.
- The multimodal task's caption/OCR index rows now set IsEnabled; only
backends with a column default were storing them as enabled.
Retrieval
- A text query scores lower against an image than against text, so when
the model embeds images the document vector search is widened by half
(capped at 500) and run at min(VectorThreshold, 0.1); each store
group's hits are then held to their own threshold before score
normalization: text at VectorThreshold, images at min(VectorThreshold,
0.1). Keyword hits on image rows are dropped, their Content being the
already-indexed caption.
- image_vector hits are searchable and merge like caption hits: parent
context is attached, and the caption is not repeated from ImageInfo.
Adds golang.org/x/image for scaling and WebP/BMP decoding.
* fix(models): report no thinking levels for embedding, rerank and ASR models
Embedding rows now carry capabilities, and the end-to-end run showed them
listing off/auto/minimal/low/medium/high: these references resolve no chat
protocol, and an empty level map reads as "every level supported". A model
without a chat protocol does not think, so its capabilities say so.
* fix(compaction): never read a cancelled turn's stream as a summary
streamSummary selects on the turn's context and the provider stream at
once. When the turn is already cancelled both are ready (providers close
the stream once their context ends), and select picks at random, so a
stopped turn could come back as a completed, partial summary with a nil
error. TestSummarizationIsNotRetriedOnceTheTurnIsCancelled caught it
intermittently in CI and reproduces under -race with several -cpu values.
Check the context before each receive and again when the stream closes.
* feat(knowledge): put image vectors behind a knowledge-base switch, off by default
Image vector indexing and recall used to turn on whenever the knowledge
base's embedding model declared image input. A model that takes images is
often chosen for text alone, so an upgrade would have added an embedding
call per image, widened vector recall 1.5x and lowered the image threshold
to 0.1 for those knowledge bases without anyone asking.
ImageProcessingConfig gains image_vector_enabled (default false; stored
JSON without the key reads as off, so existing knowledge bases stay off
with no migration). Both sides now need the switch and the model:
- indexImageVector returns before touching the model unless the KB opted
in (KnowledgeBase.IsImageVectorEnabled, which also needs vector indexing).
- applyImageRecall resolves the embedding model only when some KB in the
search opted in, and records the opted-in KBs per store group
(storeGroup.ImageKBIDs). Only those groups widen the pool and lower the
threshold; image hits of any other KB, including ones indexed before the
switch was turned off, are dropped.
The KB editor shows the switch in the image processing section, with a
warning when the selected embedding model does not declare image input;
docs describe the switch, its cost and what turning it off does.
* fix(knowledge): leave image vectors behind when cloning into a KB that is off
CloneChunk copied every chunk of the source document and then CopyIndices
for all of them, so copying a document from a knowledge base with image
vectors on into one with the switch off (same model and store) still
created image_vector chunks and their index rows, bypassing the switch
that indexImageVector honours.
Image vector chunks are now dropped from the copy when the target has not
opted in. Every engine's CopyIndices copies only the rows of chunks in the
mapping, so their index rows stay behind too.
* fix(search): refill the vector pool past stale image vectors
Image hits of a knowledge base whose switch is off were dropped only after
the engine had ranked and cut its results at TopK. Image vectors indexed
before the switch was turned off could fill that pool and push out valid
text hits: with MatchCount 5 the pool is 50, and 50 stale image rows
ranked above the text left the search empty.
No engine can filter rows by source type (Postgres and SQLite do not even
take ExcludeChunkIDs), so the exclusion cannot move into the query. When
the document vector pool comes back full and holds stale image rows, that
search alone runs again on a pool twice as large, until it holds TopK rows
that are not stale, the index runs out, or the pool reaches
maxRetrievalPoolSize (at most four extra calls from a pool of 50). The pool
is then cut back to TopK such rows, matching an engine-side exclusion
unless more than maxRetrievalPoolSize-TopK stale rows rank above the text.
* feat(i18n): translate the image vector switch and image input keys into zh-TW
The zh-TW locale arrived from main after this branch added five keys: the
knowledge base's image vector switch (label, description, unsupported-model
warning) and the embedding model's image input flag (label, description).
* fix(knowledge): leave image vectors behind when moving into a KB that is off
A reuse_vectors move relocates a document's chunks and index rows in place:
MoveKnowledgeIndices rebinds the rows by document ID and
MoveChunksByKnowledgeID rebinds the chunks. Moving a document from a
knowledge base with image vectors on into one with the switch off therefore
carried its image_vector chunks and their vectors along, the same bypass the
clone path had.
When the target has not opted in, the image_vector chunks are now deleted
during the move, their index rows first and then the chunks, so a retry
finds the chunks again rather than leaving orphaned rows. A target that
opted in still receives them unchanged.
* fix(retriever): keep image vectors out of keyword-only engines
BatchIndexVectors handed image rows to every engine of the composite,
including one that serves keyword retrieval alone (Elasticsearch v7 next to
a vector store). There the row holds nothing but the image's caption, which
the caption's own chunk already indexes, and retrieval discards every
keyword hit on it.
Image rows now go only to engines that serve vector retrieval. An engine
that serves both keeps one row for both kinds, so its keyword search still
meets them; the search side handles that.
* fix(search): refill the keyword pool past image rows
Every engine that serves both kinds of retrieval stores one row per index
entry and runs its keyword search over that row's content, so an image
vector row, whose content is its caption, is a keyword hit wherever the
caption's own chunk is. filterImageHits drops all of them, but only after
the engine has cut its ranking at TopK, so they crowded out text and caption
hits; a keyword ranking that put the image rows first came back empty.
The keyword pool is now refilled the way the vector pool is refilled past
stale image vectors: when it came back full and held image rows, the
keyword search runs again on a pool twice as large, up to
maxRetrievalPoolSize, and is cut back to TopK hits that are kept.
Leaving the rows out of the keyword index instead would mean writing them
without searchable content (Milvus and Tencent VectorDB derive a BM25
sparse vector from it, and how they take an empty one is unverified) or a
source-type filter in every engine's keyword query (ten engines, and
filterable fields that existing collections lack).
* revert: let image vectors move with the document into a KB that is off
This reverts commit e259977fe915f7d0bc30c4b02bc7af04da1c8dd4.
Deleting the image vector index rows by chunk ID before the move is not
reliable: the Weaviate and Elasticsearch v8 delete wrappers ignore partial
failures, and OpenSearch rejects more than 1000 IDs with ErrBatchTooLarge.
A partial delete followed by the chunk delete leaves index rows whose chunks
are gone, which nothing ever cleans up.
Without it, the image_vector chunks and their index rows move together
with the document. While the target's switch is off they are filtered at
query time (the pool refill keeps them from crowding out text), they are
removed with the document on delete, reparse or knowledge base delete, both
chunks and index rows by knowledge ID, and they become active again if the
target turns the switch on. The clone path keeps dropping them, since there
the rows are copied rather than moved.
* fix(search): keep the image row refill from losing hits it already had
refillPool replaced a pool with the answer to each larger search. A refill
that answered only in part (RetrieveResult.Error, swallowed into a warning
by retainPartialResults, as when some collections of the store did not
answer) could hold fewer text rows than the complete smaller pool, or none:
40 stale image rows above 10 text came back from a partial refill as the
40 image rows alone, and the search lost all 10 text hits. A refill that
failed outright failed the whole store group, discarding a first round that
had succeeded.
A refill is now best effort. A failed or incomplete refill search stops
the refill and leaves the last complete pool in place, cut back to TopK
kept rows as before, so a refill can no longer leave a search worse off
than it was without it. Merging the partial answer into the pool was not
done: its rows would need deduplicating and its ranking is cut at a
different TopK over a different set of collections, while keeping the
previous pool is exact.
* docs(knowledge): say what happens to image vectors in a KB that is off and what the refill costs
The image vector section now says that with the switch off, image vectors
that were indexed before it was turned off, or that came along when the
document was moved in from a KB that has it on, are kept but filtered at
query time, are removed when the document is deleted or reparsed or the
KB is deleted, and are recalled again once the switch is turned on; a copy
into such a KB leaves them behind.
It also states the cost of keeping dropped image rows from crowding text
out of the pools: the vector and keyword pools are searched again with
TopK doubled up to the 500 candidate cap, which adds queries for store
groups that hold such rows, and text can still be missed when more than
about 500 candidates are image rows. A failed or partial refill keeps the
previous complete pool.
* fix(search): log unexpected refill result sets, and describe the refill stop condition precisely
* fix(embedding): stop four embedders from returning (nil, nil) on a dead upstream
doRequestWithRetry keeps resp and err in the enclosing scope and returns
(nil, err) once the retries run out. Writing `req, err := ...` inside the
loop introduces a loop-local err, so `resp, err = client.Do(req)` updates
that copy and the enclosing err stays nil. Every retry failing therefore
returned (nil, nil): BatchEmbed's `if err != nil` did not fire and the next
line dereferenced resp.Body, a nil-pointer panic that took the process down
instead of surfacing "connection refused". Stopping the embedding upstream
and asking any RAG question reproduced it.
The bug was found, documented and fixed in openai.go, then re-fixed in
zhipu.go and again, differently, in azure_openai.go; jina.go, nvidia.go,
aliyun.go and volcengine.go kept the original spelling because the retry
loop exists in eight near-identical copies and a fix to one reaches none of
the others.
TestEmbeddersReportUnreachableUpstream now drives all eight against a
refused port with the retry budget zeroed, so the next copy of this loop is
covered the moment it appears.
* fix(embedding): read the documented index field of DashScope multimodal vectors
The multimodal embedding response carries each vector's position in `index`:
{"output": {"embeddings": [{"index": 0, "embedding": [...], "type": "text"},
{"index": 1, "embedding": [...], "type": "image"}]}}
The decoder read `text_index`, which is the field name of DashScope's *text*
embedding API one page over. A missing field decodes as 0, so every vector in
a batch was written to slot 0 and the rest of the slice stayed empty: with the
default BATCH_EMBED_SIZE of 5, four of every five chunks were indexed with an
empty vector. Only models routed to this client are affected — an Aliyun
embedding model whose name contains "vision" or "multimodal"; the text-only
models go through the OpenAI-compatible path.
An index outside the batch, or a response that covers fewer inputs than were
sent, is now an error. Handing back an empty vector stores it and poisons
retrieval without anything failing.
The regression test replays the response example from the reference page, so
the decoder is pinned against the vendor's own document rather than against
our structs.
https://help.aliyun.com/zh/model-studio/multimodal-embedding-api-reference
Reject out-of-range and duplicate response indexes before mapping them to batch inputs. Return an error when a requested input has no response item instead of silently passing an incomplete batch downstream.
Add focused coverage for malformed responses and valid out-of-order results.
Signed-off-by: PP1 <74917296+pengpengyi92@users.noreply.github.com>
* fix(security): harden outbound URL SSRF protection
* test(security): add SSRF validation tests for outbound URL caching
Introduce tests for outbound SSRF validation cache to ensure proper handling of URL validation, including cache reuse for the same origin, cache invalidation on whitelist changes, and expiration of cached entries. This enhances the robustness of SSRF protection mechanisms in the application.
* refactor: clean up whitespace in engine_factory.go and ssrf_outbound_cache.go
Removed unnecessary blank lines in engine_factory.go and adjusted variable declarations for consistency in ssrf_outbound_cache.go. This improves code readability without altering functionality.
If the embedding provider returns an empty slice of vectors (e.g. length 0)
due to api mismatch or empty input, BatchEmbed will return length 0.
Accessing embedding[i] directly in BatchEmbedWithPool results in an out of bounds panic.
Because the panic occurs inside a mu.Lock() block, the mutex is never unlocked,
leading to a permanent deadlock of the document processing queue.
This fix checks the returned embeddings slice length and returns an error
instead of panicking, releasing the mutex properly.
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.
Surface per-model active/waiting/limit from the background task limiter so admins can distinguish task scheduling from model-service throttling, with full i18n coverage.
Introduce a new `max_concurrency` parameter to the model configuration, allowing control over the maximum number of concurrent background tasks (ingestion/enrichment) per model. This feature is applicable to chat, embedding, and VLM models, with a value of 0 or undefined falling back to the global default. Update UI components and localization files to support this new setting, ensuring a seamless user experience.
Extend the per-model background concurrency governor to embedding, the
highest-volume background model call (document ingestion vectorises every
chunk, so a batch upload can burst the whole worker pool against one provider).
The wrapper sits innermost (below debug/langfuse) because BatchEmbedWithPool
fans a batch out into per-sub-batch BatchEmbed calls via the pooler, and the
pooler invokes BatchEmbed on whichever Embedder is threaded down as `model`;
sitting innermost routes those per-sub-batch provider round-trips through Gate
so the semaphore bounds real concurrent calls rather than one coarse per-doc
unit. Only background (asynq) calls are throttled; interactive query embedding
is never gated.
Align embedding outbound requests with chat module defenses by validating
base URLs at construction time and using SSRFSafeDialContext with redirect checks.
Reproduce:
1. Run WeKnora pointing OPENAI_BASE_URL at any OpenAI-compatible
embedding endpoint.
2. Stop that endpoint.
3. Issue any RAG-mode chat / knowledge_search query — the pipeline
reaches chunk_search_parallel → knowledgeBaseService.GetQueryEmbedding
→ OpenAIEmbedder.Embed → OpenAIEmbedder.BatchEmbed.
4. Observe the entire WeKnora process SIGSEGV instead of returning a
graceful 5xx.
Stack (real, observed in production):
panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x2 addr=0x40]
internal/models/embedding/openai.go:195 +0x458 BatchEmbed
internal/models/embedding/openai.go:93 +0xb0 Embed
internal/application/service/knowledgebase_search.go:35
GetQueryEmbedding
internal/application/service/chat_pipeline/search.go:371
searchByTargets.func1
Root cause — variable shadowing in doRequestWithRetry:
func (e *OpenAIEmbedder) doRequestWithRetry(...) (*http.Response, error) {
var resp *http.Response
var err error // outer err
...
for i := 0; i <= e.maxRetries; i++ {
...
req, err := http.NewRequestWithContext(...) // ← `:=` shadows
// outer `err`
// because `req`
// is new in
// the loop scope
if err != nil { ... continue }
...
resp, err = e.httpClient.Do(req) // writes to the
// shadowed
// loop-local err
if err == nil { return resp, nil }
logger.Errorf("attempt %d failed: %v", i+1, err) // logs the inner err
}
return nil, err // returns OUTER
// err — never set
}
When httpClient.Do returns connection errors on every attempt, the loop
logs them faithfully but the outer err stays nil. The function returns
(nil, nil). Caller in BatchEmbed reads:
resp, err := e.doRequestWithRetry(ctx, jsonData)
if err != nil { return nil, ... } // skipped (err is nil)
if resp.Body != nil { // ← line 195 · resp itself is nil
defer resp.Body.Close() // nil pointer dereference
} // → SIGSEGV → process dies
Fix — declare `req` separately so the shared `err` variable is not shadowed:
var req *http.Request
req, err = http.NewRequestWithContext(...) // outer err
After the fix, when all retries fail, doRequestWithRetry returns the real
error (e.g. `dial tcp 127.0.0.1:11434: connect: connection refused`),
the caller propagates it as a regular error, the chat pipeline returns a
5xx to the client, and the process keeps running.
Verified by stopping the embedding upstream while WeKnora was serving
queries · process stayed up · returned graceful errors · resumed normal
operation when the upstream came back. Without the patch the same
sequence killed the process within one query.
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.
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.
- 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.
Remove the hardcoded TruncatePromptTokens: 512 limit from WeKnoraCloud
BatchEmbed to allow the API to use its default truncation behavior,
avoiding unexpected truncation of longer input texts.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Remove infrastructure/crypto package (CryptoService, config, PBKDF2-based encryption)
- Replace with lightweight utils.GetAESKey/EncryptAESGCM/DecryptAESGCM helpers
- Remove cryptoSvc dependency from modelService, tenantService, weKnoraCloudService
- Remove crypto state persistence logic from container initialization
- Add TruncatePromptTokens field to WeKnoraCloud embed request
- Update frontend hint text for WeKnoraCloud credential errors
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Prevents 400 Bad Request errors when uploading documents with large
tables or code blocks by adding hard limits in buildUnitsWithProtection
and mergeUnits functions.
- Added support for tracking the status of FAQ import tasks, including progress indicators for ongoing imports.
- Introduced new localization strings for processing tasks and batch operations in multiple languages.
- Updated the FAQ entry manager to display import progress and improved user feedback during the import process.
- Enhanced the KnowledgeBase component to reflect the processing state of knowledge items, improving user experience during document uploads.
- Implemented batch deletion functionality for chunks, streamlining the management of knowledge base content.
- Updated agent configuration to support separate system prompts for web search enabled and disabled states.
- Removed deprecated agent configuration parameters to streamline settings management.
- Enhanced UI components in AgentSettings.vue to allow configuration of custom prompts based on web search status.
- Improved localization in English, Russian, and Chinese for new prompt settings and UI elements.
- Refactored related API and service logic to accommodate changes in agent configuration structure.