### What problem does this PR solve?
Browser parsed sys.query from prompts but never called set_input_value,
so node_finished inputs displayed null in the agent orchestration run
log.
Additionally, Browser’s tenant-model path could trigger unsupported
structured-output modes (response_format/tool_choice) for some
OpenAI-compatible providers (notably DeepSeek thinking models), causing
step failures.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: Cursor <cursoragent@cursor.com>
### What problem does this PR solve?
Follow-up to #15393. After #15393 fixed the OpenSearch `search()`
signature and
the doc-meta mapping, document metadata still renders as **"0 fields"**
for every
document on the OpenSearch backend (`DOC_ENGINE=opensearch`).
**Root cause.** `OSConnection.insert()` pops `id` out of the document
before
indexing:
meta_id = d_copy.pop("id", "") # id used as _id, then DROPPED from
_source
so the stored `_source` never contains an `id` field. But the doc-meta
read path
filters and sorts on that field:
- `DocMetadataService.get_metadata_for_documents()` builds
`condition = {"kb_id": kb_id, "id": doc_ids}` -> `OSConnection.search()`
emits
`Q("terms", id=doc_ids)` (a term query on the `id` field), and
- `_search_metadata()` sorts with `order_by.asc("id")`.
With `id` absent from `_source`, the terms filter matches nothing, so
`get_metadata_for_documents()` returns an empty map and the UI shows "0
fields"
-- even though the metadata was written correctly (it is visible via a
kb_id-only query).
`ESConnection.insert()` already keeps `id` (`d_copy.get("id", "")`) with
the
comment *"also keep 'id' as a regular field for sorting"*. This is a
plain
OpenSearch-only divergence (`pop()` vs `get()`).
### Fix
Mirror Elasticsearch: use `get("id")` instead of `pop("id")` so `id`
survives in
`_source`. The doc-meta mapping already declares `id` as `keyword`, so
the field
is searchable/sortable once populated.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### Affected backends
OpenSearch only. Elasticsearch already keeps `id`; Infinity / OceanBase
unaffected.
### How to reproduce
1. `DOC_ENGINE=opensearch`, create a KB, upload/parse a document, set
metadata.
2. Open the document list -> every document shows "0 fields" (the
metadata exists
in the `ragflow_doc_meta_*` index but its `_source` has no `id` field).
### Risk & backward compatibility
`insert()` is shared with the main chunk index; keeping `id` in
`_source` brings
OpenSearch in line with Elasticsearch (which already does this), so it
is parity,
not new behavior. No default / ES / Infinity / OceanBase behavior
change.
Note: affects new inserts only. Existing `ragflow_doc_meta_*` indices
created
before this change have no `id` in `_source`; re-sync metadata, or
backfill once
with `_update_by_query` (`ctx._source.id = ctx._id`).
### Test plan
- [ ] OpenSearch: after the fix the document list shows correct metadata
field
counts (not "0 fields"); metadata filter/sort by id works.
- [ ] Elasticsearch regression: unchanged.
## Summary
Closes#15720
`FulltextQueryer.paragraph` normalized its `content_tks` token string
with `[c.strip() for c in content_tks.strip() ...]`, which iterates the
string **character by character** — `"machine learning model"` becomes
20 single characters instead of 3 tokens. Those single chars are fed to
`tw.weights(..., preprocess=False)`, producing meaningless term weights
and a garbage `MatchTextExpr`.
`paragraph()` backs `Dealer.tag_content` (the KB auto-tagging feature),
so tag retrieval/scoring is silently broken for tag-enabled knowledge
bases. Every other method in this file tokenizes with `.split()` — this
is a `.strip()`-vs-`.split()` typo.
## Change
- `rag/nlp/query.py` — change `content_tks.strip()` to
`content_tks.split()` in the `paragraph` token-normalization line.
## Why it's safe
- The caller passes a space-separated token string; `.split()` recovers
the real tokens, matching the contract of `tw.weights` and the
`.split()` tokenization used by the sibling methods (`similarity`,
`question`).
- No behavior depends on the per-character expansion.
## Verification
- `python -m py_compile rag/nlp/query.py` — OK.
- Demonstrated: `"machine learning model"` → 20 single-character entries
before, 3 real tokens after. No test references `paragraph`.
Co-authored-by: seekmistar01 <seekmistar01@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
### What problem does this PR solve?
Fixes the OpenSearch side of #10747: hybrid search drops the keyword
(BM25) leg and
ends up doing plain vector search.
When a search has both a text and a vector leg, `OSConnection.search()`
throws the text
query away:
del q["query"]
q["query"] = {"knn": knn_query}
The text clause only stays on as a filter inside the knn query, so it
narrows the
candidate set but doesn't count towards scoring. So hybrid search on
OpenSearch behaves
like plain vector search, unlike the Elasticsearch backend.
What I changed:
- when both legs are present, send a real hybrid query
`{"hybrid": {"queries": [bm25, {"knn": ...}]}}` and let a
normalization-processor
search pipeline score and combine the two legs
- only the actual filters (kb_id, available_int, ...) go in the knn
filter, not the
text must clause
- create the pipeline on startup if it's missing, so there's no separate
provisioning
step. name and weights can be set under `os:` in service_conf.yaml, or
via
`OS_HYBRID_PIPELINE`; defaults are `ragflow_hybrid_pipeline` and `[0.5,
0.5]`
- normalization-processor needs OpenSearch 2.10+. on older clusters, or
when the
pipeline can't be created, log a warning and fall back to vector-only
instead of
pointing at a pipeline that doesn't exist
This is only the hybrid-search fix; `create_doc_meta_idx` is already on
main.
Testing (there's no OpenSearch path in CI): added a unit test
(`test/unit_test/rag/utils/test_opensearch_hybrid_search.py`, no
services needed) that
checks the query built in each case — hybrid + pipeline param for
text+vector, plain knn
for vector-only, plain bool for text-only, the knn filter never carrying
the text
query_string, and the vector-only fallback when the pipeline isn't
available. Also ran
it against a real OpenSearch 2.19.1 container with a doc that matches
the keyword but
sits outside the knn top-k: pure knn returns `['D1','D2','D5']` (keyword
doc missing),
the hybrid query returns `['A','D1','D2','D5']` (keyword doc present).
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
Signed-off-by: Danut Matei <matei.danut.dm@gmail.com>
### What problem does this PR solve?
Closes#15428
The hybrid score in `rag/nlp/search.py` (`rerank_by_model`) blends
reranker similarity with token similarity on a fixed `[0, 1]` scale:
```python
return tkweight * np.array(tksim) + vtweight * vtsim + rank_fea # tkweight=0.3, vtweight=0.7
```
The reranker implementations did not agree on that scale. Only three of
roughly 17 providers normalized their output, and `NvidiaRerank`
returned raw, unbounded logits. Weighted at `0.7`, a negative logit
could push a genuinely relevant chunk below pure keyword matches, and
its magnitude swamped `tksim`, which lives in `[0, 1]`. The practical
effect was that the same query produced differently scaled scores
depending on the configured reranker, and logit based providers degraded
retrieval quality instead of improving it.
This PR enforces a single scoring contract in one place:
- `Base.similarity` is now the only public entry point. It
short-circuits empty input and guarantees a normalized result. Each
provider implements its raw scoring in `_compute_rank`, which removes
sixteen duplicated empty input guards and the three scattered
normalization calls.
- Normalization is range aware. Providers that already return calibrated
`[0, 1]` relevance scores (Cohere, Jina, Voyage, and others) keep their
absolute magnitudes, so `similarity_threshold` filtering and the
reported `vector_similarity` stay meaningful. Only out-of-range output
such as NVIDIA logits is min-max rescaled into `[0, 1]`.
- The twelve leftover `[DEBUG ...]` prints in `rerank_by_model`,
introduced in #14231, are removed. They ran on every retrieval, added
per chunk overhead, and leaked queries, keywords, and document content
to stdout and logs.
A new regression suite in
`test/unit_test/rag/llm/test_rerank_normalization.py` covers logit
rescaling (positive, negative, and flat batches), preservation of
already calibrated scores, ordering, empty input handling, and the per
provider HTTP path. It also asserts that no provider overrides
`similarity()`, so the contract cannot silently drift.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Closes#15433
Reranked retrieval drops results and returns short pages once pagination
crosses the first candidate block, for the common page sizes 10 and 30.
In `rag/nlp/search.py`, the candidate window (`RERANK_LIMIT`) is rounded
up to a multiple of `page_size` to keep block based pagination aligned,
and then clamped back to 64:
```python
RERANK_LIMIT = math.ceil(64 / page_size) * page_size if page_size > 1 else 1 # e.g. 70 for page_size=10
RERANK_LIMIT = max(30, RERANK_LIMIT)
if rerank_mdl and top > 0:
RERANK_LIMIT = min(RERANK_LIMIT, top, 64) # clamps back to 64, breaking the multiple
```
`RERANK_LIMIT` is used both as the backend block size (`page =
global_offset // RERANK_LIMIT`) and as the modulus that slices a page
out of a reranked block (`begin = global_offset % RERANK_LIMIT`). When
it stops being a multiple of `page_size`, the block that gets fetched
and the slice taken from it no longer agree. With `page_size=10` and
`top=1024`, page 7 returns only 4 of 10 results and the head of the next
block is never shown on any page. This happens whenever the result set
spans more than one block, which is the default.
**Fix**
The window math is moved into a small reusable helper,
`Dealer._rerank_window`, which:
- targets a pool of about 64 candidates,
- bounds it by `top` when a reranker is active, and
- always rounds to a whole number of pages, so the window stays an exact
multiple of `page_size`.
The call site becomes a single line, and the alignment invariant now
lives in one documented place. Behavior is unchanged on every path that
was already aligned (the non reranked path and any `top` that already
produced a page multiple).
**Verification**
A simulation of the full retrieval path (per block rerank, similarity
threshold filter, and the exact `page // window` and `offset % window`
math) confirms the fix loses nothing where the old code lost real
results:
```
ps=10 top=1024: new window=70 dropped_valid=0 | old window=64 dropped_valid=16
ps=30 top=1024: new window=90 dropped_valid=0 | old window=64 dropped_valid=66
```
New unit tests in `test/unit_test/rag/test_search_pagination.py` cover
the alignment invariant, cross block pagination (every candidate
surfaced once, in order, no gaps, no short interior pages), the reported
regression, and parity with the old window on the previously correct
paths. All 114 cases pass and `ruff check` is clean.
Fixes the reranked deep pagination data loss described above.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Feat：
- Get model list from remote provider.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Closes#15461.
RAGFlow had no way to ingest Salesforce CRM data, so support / sales
teams couldn't ground responses on live Accounts, Contacts,
Opportunities, Cases, or Knowledge articles. This adds a first-class
Salesforce data source connector that authenticates against a Connected
App via OAuth 2.0 client-credentials, queries selected SObjects via
SOQL, and turns each record into an indexable document with incremental
sync.
**Highlights**
- `common/data_source/salesforce_connector.py`: new
`SalesforceConnector` (`CheckpointedConnectorWithPermSync` +
`SlimConnectorWithPermSync`).
- OAuth 2.0 client-credentials flow; canonical `instance_url` from the
token response so multi-pod orgs route correctly.
- Per-object `SystemModstamp` cursor stored in
`SalesforceCheckpoint.cursors` — a failure mid-object doesn't rewind
sibling objects, and re-syncs only fetch changed rows.
- Deterministic record-to-text formatter (sorted keys) so SOQL field
reordering on the server doesn't mark every row "changed" on each poll.
- `_get_json` raises on non-2xx so 429 / 5xx never silently advance the
checkpoint past missing data.
- `Knowledge__kav` is in the default object set but is skipped silently
when the org doesn't have Salesforce Knowledge enabled (404 on
describe).
- Slim-doc IDs are scoped as `<Object>/<Id>` so prune deletes can't
collide across object types.
- `common/constants.py`, `common/data_source/config.py`,
`common/data_source/__init__.py`: register `salesforce` in `FileSource`
/ `DocumentSource` and export `SalesforceConnector`.
- `rag/svr/sync_data_source.py`: new `Salesforce(SyncBase)` class routed
through `load_from_checkpoint` (poll_source would re-walk every object
each run) and added to `func_factory`.
- Frontend:
- `web/src/pages/user-setting/data-source/constant/index.tsx`: new
`DataSourceKey.SALESFORCE`, form fields (instance URL, client ID/secret,
objects, api_version, batch size), `syncDeletedFiles` capability,
default form values, and tile entry with the new icon.
- `web/src/locales/{en,zh}.ts`: description + per-field tooltips.
- `web/src/assets/svg/data-source/salesforce.svg`: 48x48 brand-style
icon to match the other Microsoft / cloud tiles.
**Verification**
- `npm run build` (vite + esbuild) passes (1m 26s).
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Fix:
- VolcEngine adapt to new api_key format
- Save dict api_key as json
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Closes#15465.
RAGFlow supports S3, Google Cloud Storage, R2, and OCI as data sources
but not Azure Blob Storage, leaving Azure users without a way to index
container objects into a knowledge base. This adds a first-class Azure
Blob Storage data-source connector — distinct from RAGFlow's existing
Azure storage *backends* (`rag/utils/azure_sas_conn.py`,
`rag/utils/azure_spn_conn.py`) which store RAGFlow's own files.
**Highlights**
- `common/data_source/azure_blob_connector.py`: new `AzureBlobConnector`
(`CheckpointedConnectorWithPermSync` + `SlimConnectorWithPermSync`).
- Uses the existing `azure-storage-blob` dependency (already in
`pyproject.toml`).
- Three auth modes, tried in order of precedence:
1. **Account key** — `account_name` + `account_key` + `container_name`.
2. **Connection string** — `connection_string` + `container_name`.
3. **SAS token** — `container_url` + `sas_token` (same shape as
`RAGFlowAzureSasBlob`).
- ETag fingerprint stored per blob in `AzureBlobCheckpoint.etags` —
unchanged blobs (same ETag as last run) are skipped without a download.
Only new/modified blobs are fetched.
- Optional `prefix` scopes indexing to a virtual folder.
- `validate_connector_settings()` probes `get_container_properties()`
and maps `AuthenticationFailed / 403 / ContainerNotFound` to typed
connector exceptions.
- Slim-doc IDs are blob names so prune reconciles correctly.
- `common/constants.py`, `common/data_source/config.py`,
`common/data_source/__init__.py`: register `azure_blob` in `FileSource`
/ `DocumentSource` and export `AzureBlobConnector`.
- `rag/svr/sync_data_source.py`: new `AzureBlob(SyncBase)` class routed
through `load_from_checkpoint` (ETag fingerprint owns change-detection)
and added to `func_factory`.
- Frontend:
- `web/src/pages/user-setting/data-source/constant/index.tsx`: new
`DataSourceKey.AZURE_BLOB`, auth-mode selector (account key / connection
string / SAS token), all credential fields, prefix + batch-size,
`syncDeletedFiles` capability, default form values, tile entry with
icon.
- `web/src/locales/{en,zh}.ts`: description + per-field tooltips for all
9 new keys.
- `web/src/assets/svg/data-source/azure-blob.svg`: Azure-branded
stacked-cylinders icon.
**Verification**
- `npm run build` (vite + esbuild) passes (37 s).
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
## Problem
When uploading `.md` files with `parser=naive` and `delimiter="\n"`,
markdown headers (e.g., `## Quick Travel`) become separate chunks with
very short content (16-18 characters). This causes retrieval issues:
when the header is matched, the corresponding body text is not included
in the chunk.
## Related Issues
Closes#15487
## Checklist
- [x] Code changes are minimal and focused
- [x] Unit tests added (12/12 passed)
- [x] No breaking changes
add the newanthropic and voyage models. Strip opus 4.7 and 4.8 of
certain usnspported keys
Co-authored-by: Idriss Sbaaoui <112825897+6ba3i@users.noreply.github.com>
### What problem does this PR solve?
This PR fixes the issue where Qwen3.5/Qwen3.6 series models may spend
excessive time on simple document-parsing tasks, such as Auto Metadata
extraction, keyword extraction, question generation, and image
description when using the MinerU parser.
For these tasks, Qwen3.5/Qwen3.6 models may perform unnecessary
reasoning by default, which can lead to very long response times, high
token consumption, and, in some cases, potential infinite output loops.
Since Qwen3.5/Qwen3.6 multimodal models are instantiated as `CvModel`
when configured as `image2text`, the existing `enable_thinking=False`
logic in `chat_model.py` does not apply to them. This PR adds the
corresponding handling for the CV/image-to-text model path as well.
This helps reduce unnecessary thinking time, avoid potential infinite
loops, and improve parsing efficiency without noticeably affecting
output quality for these simple extraction and image-description tasks.
Fixes#15083.
### What problem does this PR solve?
Fixes#15117.
Chunk images are stored with `img_id = f"{bucket}-{objname}"` in
`image2id()` (`rag/utils/base64_image.py`). When loading via
`id2image()`, the code used `image_id.split("-")` and required exactly
two segments. Object keys that contain hyphens (e.g. `page-1.jpg`)
produce more than two segments, so `id2image` returns `None` and chunk
image previews fail even though the blob exists.
This is the same parsing issue as #15115 (HTTP thumbnail route); this PR
fixes the indexing/retrieval path.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [ ] New Feature (non-breaking change which adds functionality)
- [ ] Documentation Update
- [ ] Refactoring
- [ ] Performance Improvement
- [ ] Other (please describe):
### Test plan
- [x] `pytest test/unit_test/rag/utils/test_base64_image.py`
- [ ] Manual: index a chunk with an `objname` containing hyphens and
confirm `img_id` resolves to an image in retrieval
Fixes#15117.
### What problem does this PR solve?
Fixes#15427.
All LiteLLM-routed chats fail with:
- Anthropic: `litellm.BadRequestError: AnthropicException -
{"type":"invalid_request_error","message":"model_type: Extra inputs are
not permitted"}`
- OpenAI: `litellm.BadRequestError: OpenAIException - Unknown parameter:
'model_type'`
This is a regression from v0.25.4.
#### Root cause
A chat assistant's `llm_setting` is forwarded to the model as
`gen_conf`. `llm_setting` can legitimately carry RAGFlow-internal
metadata such as `model_type` (the chat REST APIs in
`api/apps/restful_apis/` read it back out of `llm_setting`), so that key
ends up inside `gen_conf`.
`Base._clean_conf` (OpenAI-compatible providers) already **whitelists**
the keys it forwards, so direct-OpenAI providers were unaffected.
`LiteLLMBase._clean_conf` only dropped `max_tokens` and passed
everything else straight through to `litellm.acompletion`, which
forwarded `model_type` to the upstream provider — and Anthropic / OpenAI
reject it. Because both Claude and GPT route through LiteLLM, every chat
broke.
#### Fix
- Extract the allowed-key set into a shared `ALLOWED_GEN_CONF_KEYS`
constant and reuse it in `Base._clean_conf`.
- Apply the same whitelist in `LiteLLMBase._clean_conf`, plus the
LiteLLM-specific reasoning params (`thinking`, `reasoning_effort`,
`extra_body`) that the model-family policies inject for reasoning
models.
This covers all four LiteLLM completion paths (`async_chat`,
`async_chat_streamly`, `async_chat_with_tools`,
`async_chat_streamly_with_tools`), since they all route through
`_clean_conf`.
#### Tests
Adds `test/unit_test/rag/llm/test_clean_conf_whitelist.py` covering both
backends: `model_type` (and other stray keys) are dropped, genuine
generation params and `thinking` survive, `max_tokens` is removed, and
the whitelist invariants hold.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] Added test cases
fix: restore TitleChunker output for json/chunks upstream formats
## Summary
The refactor commit e194027b (#14247) introduced two regressions that
caused `TitleChunker` to produce zero chunks when the upstream Parser
node outputs `json` or `chunks` format (e.g. PDF parsing).
## Root Cause
### 1. Dead code in `extract_line_records` (critical)
After refactor, when `payload` is `None` (which is the case for `json`
and `chunks` output formats), the method returns an empty list
immediately via `return []`, so no records are ever extracted from
structured upstream output. The original `json`/`chunks` handling code
became unreachable dead code.
### 2. Unconditional overwrite in `build_chunks_from_record_groups`
The `chunks` variable assigned in the `if` branch for markdown/text/html
formats was unconditionally overwritten by the statement below it, due
to a missing `else` keyword.
## Fix
- Remove the premature `return []` so the `json`/`chunks` branch is
reachable again.
- Add `else` branch in `build_chunks_from_record_groups` so the two
format families are handled independently.
## Test Plan
- [x] Verified no lint errors on the changed file
- [ ] Tested with a PDF document parsed via DeepDOC → TitleChunker
pipeline
- [ ] Tested with markdown input through TitleChunker
- [ ] Tested hierarchy and group chunking modes
## Impact
- Fixes the regression where documents parsed with `json`/`chunks`
output format produced no chunks from `TitleChunker`.
- No API or configuration changes. Fully backward compatible.
Signed-off-by: noob <yixiao121314@outlook.com>
### What problem does this PR solve?
Closes#15332.
RAGFlow can index Gmail and generic IMAP mailboxes but had no native
connector for Outlook / Microsoft 365 mail. Organisations on Microsoft
365 had no way to bring mailbox content into a knowledge base through
Microsoft Graph.
This PR adds a net-new Outlook data source that:
- Authenticates against Microsoft Graph with the same MSAL
client-credentials flow already used by the SharePoint and Teams
connectors (no new auth primitives).
- Pages over `/users/{id}/mailFolders/{folder}/messages/delta` per
mailbox and persists `@odata.deltaLink` values in
`OutlookCheckpoint.delta_links`, so incremental syncs only fetch changed
messages.
- Supports two scoping modes:
- **Tenant-wide** (default): enumerates every user in the tenant via
`/users` and syncs each mailbox. Requires `User.Read.All`.
- **Targeted**: when `user_ids` is provided (comma-separated UPNs or
object IDs), only those mailboxes are synced. `User.Read.All` is not
needed in this mode.
- Lets the caller pick the mail folder (`inbox`, `sentitems`, `archive`,
...). Defaults to `inbox`.
- Maps each message to a `Document` shaped after the Gmail connector:
one `TextSection` carrying `From/To/Cc/Subject` headers + body, with
HTML bodies stripped to text inline (no extra dependency).
- Surfaces typed errors on the validation probe:
401 → `ConnectorMissingCredentialError`, 403 →
`InsufficientPermissionsError` (with `Mail.Read` / `User.Read.All`
hint), 404 on a configured mailbox → `ConnectorValidationError`, 5xx →
`UnexpectedValidationError`.
- Skips messages flagged `@removed` by the delta semantics and messages
whose `receivedDateTime` is older than `poll_range_start`.
#### Files
| File | Change |
|------|--------|
| `common/data_source/outlook_connector.py` | **New** —
`OutlookConnector` (`CheckpointedConnectorWithPermSync` +
`SlimConnectorWithPermSync`) + `OutlookCheckpoint` + tiny `_strip_html`
helper. |
| `common/data_source/config.py` | `DocumentSource.OUTLOOK = "outlook"`.
|
| `common/constants.py` | `FileSource.OUTLOOK = "outlook"`. |
| `common/data_source/__init__.py` | Export `OutlookConnector`. |
| `rag/svr/sync_data_source.py` | `Outlook(SyncBase)` with `batch_size`
normalisation, CSV/list parsing of `user_ids`; registered in
`func_factory`. |
| `web/src/pages/user-setting/data-source/constant/index.tsx` |
`DataSourceKey.OUTLOOK`, visibility map (`syncDeletedFiles: true`), info
entry, form fields (tenant_id, client_id, client_secret, folder,
user_ids, batch_size), default values. |
| `web/src/locales/en.ts`, `web/src/locales/zh.ts` |
`outlookDescription` + 5 tooltip keys (EN + ZH). |
| `test/unit_test/data_source/test_outlook_connector_unit.py` | **New**
— 19 unit tests (`p1`/`p2`/`p3`) covering auth, validation (tenant-wide
vs specific user vs error paths), checkpoint helpers, user enumeration
pagination, message filtering, HTML body stripping. |
#### Required Azure AD permissions
- `Mail.Read` (Application, admin-granted) — always.
- `User.Read.All` (Application, admin-granted) — only when `user_ids` is
left blank so the connector can enumerate mailboxes.
#### Out of scope
- **Attachment indexing.** The current connector emits message body +
headers; binary attachments are flagged via `metadata.has_attachments`
but not pulled. Adding attachment hydration is straightforward but
scoped out per the issue's "decide whether attachments are indexed in
the first version" note.
- **Delegated (per-user) OAuth.** The connector uses app-only
credentials, consistent with the SharePoint / Teams precedent in this
codebase.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Document metadata is completely broken on the OpenSearch backend
(`DOC_ENGINE=opensearch`). Both failures were introduced by #14577,
which added
a doc-metadata dispatch surface but only validated it against
Elasticsearch.
**1. Index creation rejected (`mapper_parsing_exception`).**
`OSConnection.create_doc_meta_idx` feeds `conf/doc_meta_es_mapping.json`
verbatim to OpenSearch. That file declares a top-level `"dynamic":
"runtime"`.
Runtime fields are Elasticsearch-only; OpenSearch cannot parse the
value:
mapper_parsing_exception: Could not convert [dynamic.dynamic] to boolean
(400)
**2. `search()` signature mismatch (`TypeError`).**
`DocMetadataService` (added by #14577) calls `docStoreConn.search(...)`
with
snake_case kwargs (`select_fields=`, `index_names=`,
`knowledgebase_ids=`, …),
matching `ESConnection.search`. But `OSConnection.search` still uses
camelCase
parameters (`selectFields`, `indexNames`, `knowledgebaseIds`, …):
TypeError: OSConnection.search() got an unexpected keyword argument
'select_fields'
The UI then shows "0 fields" for every document on OpenSearch.
### Fix
1. In `OSConnection.create_doc_meta_idx`, normalize a top-level
`"dynamic": "runtime"` to `True` **for the OpenSearch request only**.
The
shared mapping file is left untouched, so the Elasticsearch backend
keeps its
runtime-field behavior. Dynamic field discovery is preserved on
OpenSearch.
2. Rename the `OSConnection.search()` parameters (and their in-method
local
uses) from camelCase to snake_case so they match `ESConnection.search()`
and
the `DocMetadataService` call sites. The change is confined to
`search()`;
`get/insert/update/delete` keep their existing positional signatures
(they
are called positionally from `rag/nlp/search.py`).
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### Affected backends
OpenSearch only. Elasticsearch, Infinity and OceanBase are untouched.
### How to reproduce
1. `DOC_ENGINE=opensearch`, restart the stack.
2. Upload/parse a document, then open the dataset's document list / set
metadata.
- Before: index creation 400s (`Could not convert [dynamic.dynamic]`),
and/or
`TypeError ... 'select_fields'`; document metadata shows 0 fields.
### Risk & backward compatibility
- ES default deployment: no change. `doc_meta_es_mapping.json` is not
modified,
so ES still receives `"dynamic": "runtime"`.
- `search()` rename is internal; the only kwarg caller
(`DocMetadataService`)
already uses the snake_case names this PR aligns to.
### Test plan
- [ ] `DOC_ENGINE=opensearch`: per-tenant `ragflow_doc_meta_*` index is
created
(no `mapper_parsing_exception`); document metadata reads/writes work.
- [ ] `DOC_ENGINE=elasticsearch` regression: doc-meta index still
created with
runtime mapping; metadata unchanged.
### What problem does this PR solve?
On the OpenSearch backend (`DOC_ENGINE=opensearch`), every retrieval
that
performs the KNN second-pass scoring crashes with:
AttributeError: 'OSConnection' object has no attribute 'get_scores'
**Root cause.** #14970 ("Refactor: Drop the vector fetch for ES") added
a
`get_scores()` helper to `ESConnectionBase`
(`common/doc_store/es_conn_base.py`)
and introduced `Dealer._knn_scores()` in `rag/nlp/search.py`, which
calls
`self.dataStore.get_scores(res)`. `search.py` routes Infinity and
OceanBase to
their own similarity paths via `DOC_ENGINE_INFINITY` /
`DOC_ENGINE_OCEANBASE`,
but OpenSearch sets neither flag, so it falls into the Elasticsearch
branch and
calls `get_scores`. `OSConnection` (which subclasses
`DocStoreConnection`
directly, not `ESConnectionBase`) never received that method, so any
vector-search hit triggers the crash. It reproduces with any normal
embedding
(e.g. 1024-dim mistral-embed) as soon as a KNN query returns hits.
### Fix
Add `OSConnection.get_scores()`, mirroring
`ESConnectionBase.get_scores()`.
OpenSearch hit headers expose `_score` exactly like Elasticsearch (the
existing
`OSConnection.__getSource` already reads `d["_score"]`), so the
implementation
is identical.
Scope note: Infinity and OceanBase deliberately do not use `get_scores`
(#14970 routes them elsewhere), so this fix is intentionally limited to
the
OpenSearch backend, which is the only one reaching the ES KNN-score
path.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### Affected backends
OpenSearch only. Elasticsearch already implements `get_scores`; Infinity
/
OceanBase are routed away from it.
### How to reproduce
1. `DOC_ENGINE=opensearch` (docker `.env`), restart the stack.
2. Create a knowledge base with any dense embedding model and parse a
document.
3. Run a retrieval / chat over that KB -> 500 with the AttributeError
above.
### Risk & backward compatibility
None for the default Elasticsearch deployment -- the change only adds a
method
to `OSConnection`. No default values or ES/Infinity/OceanBase behavior
change.
### Test plan
- [ ] With `DOC_ENGINE=opensearch`, retrieval over a KB returns scored
chunks
(no AttributeError).
- [ ] `DOC_ENGINE=elasticsearch` regression: retrieval unchanged.
- [ ] Empty-result path: `_knn_scores` early-returns `{}` (guarded),
get_scores
handles an empty `hits` list gracefully.
### What problem does this PR solve?
Added 4 new models:
deepseek-ai/DeepSeek-V4-Pro
deepseek-ai/DeepSeek-V4-Flash
Pro/moonshotai/Kimi-K2.6
Pro/zai-org/GLM-5.1
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Closes#15330.
RAGFlow had no connector for OneDrive / OneDrive for Business. Users who
store working documents in OneDrive could not index them into a
knowledge base without manually downloading and re-uploading files.
This PR adds a net-new OneDrive data source that:
- Authenticates against Microsoft Graph with the same MSAL
client-credentials flow already used by the SharePoint and Teams
connectors (no new auth primitives).
- Enumerates every drive visible to the service principal and pages
through `/drives/{id}/root/delta`, persisting `@odata.deltaLink` values
per drive so subsequent syncs only fetch changed items.
- Optionally narrows ingestion to a sub-folder (`folder_path`) without
needing a separate code path.
- Surfaces typed errors on the validation probe (`GET /drives?$top=1`):
401 → `ConnectorMissingCredentialError`, 403 →
`InsufficientPermissionsError` (with a `Files.Read.All` hint), 5xx →
`UnexpectedValidationError`.
- Filters folders, soft-deleted items, and unsupported extensions (`.pdf
.docx .doc .xlsx .xls .pptx .ppt .txt .md .csv`).
#### Files
| File | Change |
|------|--------|
| `common/data_source/onedrive_connector.py` | **New** —
`OneDriveConnector` + `OneDriveCheckpoint`. |
| `common/data_source/config.py` | `DocumentSource.ONEDRIVE =
"onedrive"`. |
| `common/constants.py` | `FileSource.ONEDRIVE = "onedrive"`. |
| `common/data_source/__init__.py` | Export `OneDriveConnector`. |
| `rag/svr/sync_data_source.py` | `OneDrive(SyncBase)` with `batch_size`
normalisation; registered in `func_factory`. |
| `web/src/pages/user-setting/data-source/constant/index.tsx` |
`DataSourceKey.ONEDRIVE`, visibility map (`syncDeletedFiles: true`),
info entry, form fields (tenant_id, client_id, client_secret,
folder_path, batch_size), default values. |
| `web/src/locales/en.ts`, `web/src/locales/zh.ts` |
`onedriveDescription` + 4 tooltip keys (EN + ZH). |
| `test/unit_test/data_source/test_onedrive_connector_unit.py` | **New**
— 13 unit tests (`p1`/`p2`) covering auth, validation, checkpoint
helpers, and document filtering. |
#### Required Azure AD permission
`Files.Read.All` (Application, admin-granted).
#### Out of scope
- Interactive end-user OAuth (delegated permissions) — the connector
uses app-only credentials, consistent with the SharePoint / Teams
precedent.
- Binary download of file contents — the sync layer emits `Document`s
carrying `webUrl` + metadata; bytes are hydrated downstream by the parse
pipeline.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Python implementation of the Go-based model_provider API suite.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
---------
Co-authored-by: bill <yibie_jingnian@163.com>
### What problem does this PR solve?
Closes#15191.
RAGFlow shipped a Microsoft Teams connector stub
(`common/data_source/teams_connector.py`) whose document-loading methods
all returned `[]`, `Teams._generate()` was a `pass`, and Teams was
commented out of the data-source settings UI. As a result there was no
way to index Teams channel conversations into a knowledge base.
This PR implements the connector end to end on top of Microsoft Graph
(Office365-REST-Python-Client). It shares the MSAL client-credentials
auth shape with the SharePoint connector.
**Backend**
- `common/data_source/teams_connector.py`
- `load_credentials()` now builds the Graph client using an MSAL
client-credentials **token callback** — the form `GraphClient` actually
expects. (The previous stub passed a raw access-token string to
`GraphClient(...)`, which is not how that client is driven.) Token
acquisition is lazy, so credential loading performs no network call.
- `validate_connector_settings()` lists teams via Graph.
- `load_from_checkpoint()` is now a generator that pages teams →
channels → messages, flattens each top-level post together with its
replies into one blob-based `Document` (`extension` `.txt`/`.html`,
`blob`, `size_bytes`, `doc_updated_at`). Incremental syncs are bounded
by message `lastModifiedDateTime` (falling back to `createdDateTime`).
Per-message errors surface as `ConnectorFailure` instead of aborting the
run.
- `retrieve_all_slim_docs_perm_sync()` yields id-only `SlimDocument`
batches and the checkpoint helpers return proper `TeamsCheckpoint`s.
- ACL → `ExternalAccess` mapping is intentionally left best-effort
(`load_from_checkpoint_with_perm_sync` delegates to the standard load)
because the sync pipeline does not currently persist `ExternalAccess`.
- `rag/svr/sync_data_source.py`
- Implemented `Teams._generate()` using the existing
`CheckpointOutputWrapper` pattern (same shape as Confluence/Jira/Google
Drive), supporting full reindex and incremental polling from
`poll_range_start`.
- `TeamsConnector` is already exported from
`common/data_source/__init__.py`.
**Frontend (`web/`)**
- Enabled the `TEAMS` data-source enum and added its form fields
(`tenant_id`, `client_id`, `client_secret`), default values, display
metadata, and a Teams icon.
- Added `teamsDescription` / `teamsTenantIdTip` to `en.ts` and `zh.ts`.
**Tests**
- `test/unit_test/data_source/test_teams_connector_unit.py`: mock-based
unit tests covering credential loading (incomplete creds raise, happy
path sets the Graph client, fetch-without-creds raises), post/reply
flattening (incl. the HTML vs text extension), incremental
`lastModifiedDateTime` filtering, and slim-doc listing. All 6 pass;
`ruff check` is clean.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Closes#15187.
RAGFlow shipped a Slack connector
(`common/data_source/slack_connector.py`) but it was never usable:
`Slack._generate()` in the sync worker was a `pass` stub, the
connector's document-generating code was incompatible with the current
data model,
and Slack was commented out of the data-source settings UI. As a result,
teams had no way to index Slack channels/threads into a knowledge base.
This PR completes the connector end to end.
**Backend**
- `common/data_source/slack_connector.py`
- Rewrote `thread_to_doc` to produce a blob-based `Document`
(`extension`/`blob`/`size_bytes`). The previous implementation built the
doc with a `sections=[...]` argument and omitted the now-required
`blob`/`extension`/ `size_bytes` fields, so it raised a validation error
against the current `Document` model. Thread messages are now cleaned
and flattened into a single UTF-8 text blob.
- Added `load_from_state()` / `poll_source(start, end)` generators. The
connector's checkpoint interface is a no-op stub, so both full and
incremental syncs run through a single channel-iterating generator built
on the existing module helpers (`get_channels`, `filter_channels`,
`get_channel_messages`, `_process_message`), with per-channel thread
de-duplication.
- `rag/svr/sync_data_source.py`
- Implemented `Slack._generate()`. Credentials are loaded via
`StaticCredentialsProvider` (the connector requires `slack_bot_token`
and does not support `load_credentials`). Supports full reindex and
incremental polling from `poll_range_start`, plus the optional channel
filter. Modeled on the Confluence/Dropbox wrappers.
- `SlackConnector` was already exported from
`common/data_source/__init__.py`.
**Frontend (`web/`)**
- Enabled the `SLACK` data-source enum and added its form fields (Slack
bot token + optional channel filter), default values, display metadata,
and a Slack icon.
- Added `slackDescription` / `slackBotTokenTip` / `slackChannelsTip`
strings to `en.ts` and `zh.ts`.
**Tests**
- `test/unit_test/data_source/test_slack_connector_unit.py`: unit tests
covering credential loading (`load_credentials` raises,
`set_credentials_provider` initializes clients, missing credentials
raises) and document generation (standalone message + flattened thread,
blob/extension/size_bytes/metadata, and the incremental poll time
window). All 5 pass; `ruff check` is clean.
Required Slack scopes: `channels:read`, `channels:history`,
`users:read`.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Closes#15189.
RAGFlow shipped a SharePoint connector stub
(`common/data_source/sharepoint_connector.py`) whose document-loading
methods all returned `[]`, `SharePoint._generate()` was a `pass`, and
SharePoint was commented out of the data-source settings UI. As a result
there was no way to index files stored in SharePoint document libraries.
This PR implements the connector end to end on top of Microsoft Graph
(Office365-REST-Python-Client).
**Backend**
- `common/data_source/sharepoint_connector.py`
- `load_credentials()` now builds the Graph client using an MSAL
client-credentials **token callback** — the form `GraphClient` actually
expects. (The previous stub passed a raw access-token string to
`GraphClient(...)`, which is not how that client is driven.) Token
acquisition is lazy, so credential loading does no network call.
- `validate_connector_settings()` resolves the configured site via
Graph.
- `load_from_checkpoint()` is now a generator that enumerates every
document library under the site, walks folders depth-first, downloads
each file, and yields blob-based `Document` objects (`extension` /
`blob` / `size_bytes` / `doc_updated_at`). Incremental syncs are bounded
by file `lastModifiedDateTime`. Per-file errors are surfaced as
`ConnectorFailure` rather than aborting the run.
- `retrieve_all_slim_docs_perm_sync()` yields id-only `SlimDocument`
batches (no downloads) and the checkpoint helpers return proper
checkpoints.
- ACL → `ExternalAccess` mapping is intentionally left best-effort
(`load_from_checkpoint_with_perm_sync` delegates to the standard load)
because the sync pipeline does not currently persist `ExternalAccess`;
this can be extended once that plumbing exists.
- `rag/svr/sync_data_source.py`
- Implemented `SharePoint._generate()` using the existing
`CheckpointOutputWrapper` pattern (same shape as Confluence/Jira/Google
Drive), supporting full reindex and incremental polling from
`poll_range_start`.
- `SharePointConnector` is already exported from
`common/data_source/__init__.py`.
**Frontend (`web/`)**
- Enabled the `SHAREPOINT` data-source enum and added its form fields
`site_url`, `tenant_id`, `client_id`, `client_secret`), default values,
display metadata, and a SharePoint icon.
- Added `sharepointDescription` / `sharepointSiteUrlTip` to `en.ts` and
`zh.ts`.
**Tests**
- `test/unit_test/data_source/test_sharepoint_connector_unit.py`:
mock-based unit tests covering credential loading (incomplete creds
raise, happy path sets the Graph client, fetch-without-creds raises),
drive traversal + file download, incremental `lastModifiedDateTime`
filtering, and slim-doc listing. All 6 pass; `ruff check` is clean.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
1. Break huge function into smaller pieces
2. Add unit test for the smaller pieces function
3. Layer-ed design
a. infra layer - task_context.py, recording_context.py,
write_operation_interceptor.py, ...
b. service layer - *_service.py
c. business layer - task_handler.py
4. Default behavior: use "refactor-ed version" - can switch to original
version by change env variable
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] Refactoring
- [x] Performance Improvement
---------
Co-authored-by: Liu An <asiro@qq.com>
Co-authored-by: Zhichang Yu <yuzhichang@gmail.com>
## Summary
Fixes the confirmed asyncio anti-patterns from #14755. Only the three
verified bugs are addressed; patterns already correctly using
`asyncio.new_event_loop()` in a fresh thread are left untouched.
### Changes
**`api/apps/restful_apis/tenant_api.py` — fire-and-forget
`send_invite_email`**
`asyncio.create_task()` was called without storing the `Task` reference.
CPython's GC can collect an unfinished task, silently cancelling it and
swallowing exceptions. Fixed by storing the task in a module-level
`_background_tasks: set[Task]` with a `done_callback` to discard it on
completion — the standard Python idiom for safe background tasks.
**`api/apps/restful_apis/agent_api.py` — fire-and-forget
`background_run`**
Same root cause in the webhook "Immediately" execution path. Same fix
applied.
**`rag/llm/chat_model.py` (`LocalLLM._stream_response`) —
`asyncio.get_event_loop()` on running loop**
`asyncio.get_event_loop()` returns Quart's running event loop when
called from an async context.
Calling `loop.run_until_complete()` on it raises `RuntimeError`.
Replaced with `asyncio.new_event_loop()` so the generator
uses a dedicated fresh loop, closed in a `finally` block.
## What was NOT changed
- `llm_service._sync_from_async_stream` and
`evaluation_service._sync_from_async_gen`: both already correctly use
`asyncio.new_event_loop()` inside a fresh thread.
- `llm_service._run_coroutine_sync`: only caller is `rag/app/resume.py`
(sync context), so `thread.join()` is correct there.
- `requests` in agent tools: sync methods dispatched through thread
pools; httpx migration is a separate, larger refactor.
## Test plan
- [ ] Invite a team member and confirm the email is sent with no task
warnings in logs.
- [ ] Trigger a webhook agent in "Immediately" mode; confirm canvas
state is persisted after background run.
- [ ] Verify `LocalLLM` (Jina backend) chat and streaming work
end-to-end.
Closes#14755
---------
Co-authored-by: Zhichang Yu <yuzhichang@gmail.com>
### What problem does this PR solve?
1. Enhance retry and timeout, and adjust the default timeout
2. NER: spacy do not batch chunks
3. extract _has_cancel_and_exit
4. enhance log messages
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
- [x] Refactoring
## Summary
Fixes 10 unguarded `response.choices[0]` accesses that cause
`IndexError` or `AttributeError` when the LLM returns an empty `choices`
list — the scenario described in #14711.
- `rag/llm/cv_model.py`
- `rag/llm/chat_model.py`
Each access site is now guarded with:
```python
if not response.choices:
raise ValueError("LLM returned empty response")
```
## Verification
Detected and verified by [pact](https://github.com/qizwiz/pact) — a
sheaf-cohomological LLM contract checker using Z3 as a local theory
solver.
**pact sheaf-cohomological proof status after fix:**
| File | Ȟ¹ (after) | Z3 |
|------|-----------|-----|
| `rag/llm/cv_model.py` | 0 | UNSAT ✓ |
| `rag/llm/chat_model.py` | 0 | UNSAT ✓ |
All access sites proven safe (Z3 UNSAT certificate).
The checker was also used to verify the autogen streaming-None fix in
[microsoft/autogen#7711](https://github.com/microsoft/autogen/pull/7711).
## Test plan
- [ ] Existing test suite passes
- [ ] Manually test with a provider that returns empty `choices` under
load (e.g. Vertex AI)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Signed-off-by: Jonathan Hill <jonathan.f.hill@gmail.com>
### What problem does this PR solve?
Fixes#14997.
RAPTOR builds on the Infinity backend have been broken since v0.25.2
introduced the `extra` field in code (`rag/svr/task_executor.py:1011`)
without declaring it in `conf/infinity_mapping.json`. Every RAPTOR job
fails with:
```
infinity.common.InfinityException: (3013, 'Fail to bind the expression: extra@src/planner/expression_binder_impl.cpp:99')
```
The auto-migration in
`common/doc_store/infinity_conn_base.py:_migrate_db()` adds any columns
it finds in the mapping JSON to existing tables — so the only thing
standing between users and a working RAPTOR build is that one missing
declaration. OceanBase, ES, and OpenSearch were unaffected because they
store `extra` as a native JSON type; only Infinity (which has a strict
`varchar`/`integer`/`float` schema) needed the addition.
### The fix
Two-part change:
1. **`conf/infinity_mapping.json`**: declare `"extra": {"type":
"varchar", "default": ""}`. On next startup, `_migrate_db()` adds the
column to all existing chunk tables — no manual DDL needed for upgrading
installations.
2. **`rag/utils/infinity_conn.py` `insert()`**: serialize the `extra`
dict to a JSON string at write time, since Infinity's `varchar` can't
store a Python dict directly. Modelled on the existing `chunk_data`
handling a few lines above.
The read path (`rag/utils/raptor_utils.py:_as_extra_dict`) already
normalises both dict and JSON-string inputs, so no read-side change is
needed. Other backends are untouched — `task_executor.py` still writes
the dict, and the OceanBase/ES/OpenSearch insert paths handle dicts
natively.
### Verification
Tested on a v0.25.4 deployment with the Infinity backend by applying the
same two changes via mounted-volume override:
- Confirmed `_migrate_db()` adds the `extra` column to all pre-existing
chunk tables on startup (column visible via Infinity's
`show_columns()`).
- Triggered RAPTOR builds on four datasets (~21k chunks total) via `POST
/api/v1/datasets/<id>/index?type=raptor`.
- All four progressed past the previously-failing
`get_raptor_chunk_methods()` call into actual entity-extraction and
clustering work without the (3013) error.
- GraphRAG builds (which can trigger the same path indirectly via
`task_executor.py:857`) also progressed cleanly.
### Type of change
- [X] Bug Fix (non-breaking change which fixes an issue)
## Summary
- Adds a lightweight `@tool` decorator and `FunctionToolSession` adapter
in `rag/llm/tool_decorator.py` that let callers register plain Python
functions as LLM tools without hand-writing OpenAI function schemas or
building an MCP-style session.
- Refactors `Base.bind_tools` and `LiteLLMBase.bind_tools` in
`rag/llm/chat_model.py` to accept either the new decorator form
`bind_tools(tools=[fn1, fn2])` or the existing `(toolcall_session,
tools_schemas)` form, so existing agent/dialog call-sites in
`agent/component/agent_with_tools.py`, `api/db/services/llm_service.py`,
and `api/db/services/dialog_service.py` are unaffected.
- Adds 8 unit tests in `test/unit_test/rag/llm/test_tool_decorator.py`
covering schema shape, required/optional inference, sync + async
dispatch, and bad-input rejection.
## Usage
```python
from rag.llm.tool_decorator import tool
@tool
def get_weather(city: str) -> str:
"""Get current weather for a city.
:param city: City name to look up.
"""
return f"{city}: 21 C, partly cloudy"
chat_mdl.bind_tools(tools=[get_weather])
ans, tk = await chat_mdl.async_chat_with_tools(system, history)
```
The decorator introspects `inspect.signature` + type hints + the
docstring (`:param name:` style) and attaches an OpenAI-format
`openai_schema` to the callable. `FunctionToolSession` duck-types the
existing `ToolCallSession` protocol, dispatching async callables
directly and sync ones through `thread_pool_exec` so the event loop is
never blocked.
## Design notes
- `tool_decorator.py` deliberately does **not** live inside
`rag/llm/__init__.py` to avoid forcing every consumer through the heavy
provider auto-discovery loop and to sidestep a circular import
(`__init__.py` imports `chat_model`, which would otherwise need symbols
from `__init__.py`).
- `FunctionToolSession` is duck-typed against
`common.mcp_tool_call_conn.ToolCallSession` rather than explicitly
inheriting from it, so importing the decorator doesn't pull the MCP
client SDK into the import graph.
- Docstring parsing is intentionally minimal (`:param name:` only) to
keep this dependency-free; Google/NumPy styles can be added later via
`docstring_parser` if needed.
## Test plan
- [x] `python -m pytest test/unit_test/rag/llm/test_tool_decorator.py
-v` — 8 passed
- [x] `python -m pytest test/unit_test/rag/llm/
--ignore=test/unit_test/rag/llm/test_perplexity_embed.py` — 11 passed
(the ignored test has a pre-existing `numpy` import that's unrelated)
- [ ] Reviewer: smoke-test the new path end-to-end with a live model via
`chat_mdl.bind_tools(tools=[my_fn])` to confirm the OpenAI-format
schemas pass through unchanged
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
### What problem does this PR solve?
Feat: add local & ssh provider in admin panel
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
## Summary
Closes#14869.
Adds VLM-based semantic descriptions to **image chunks produced by the
MinerU parser**, closing a long-standing parity gap with the deepdoc
parser's `VisionFigureParser`. A maintainer flagged this in #13342
("We may add the VLM enhancement to MinerU parser as well") and an
earlier proposal exists in #13824; this PR lands the change end-to-end
inside the existing parser plumbing.
## Why
Today the MinerU parser returns image chunks containing only the
native `image_caption` and `image_footnote` strings from MinerU's
JSON. When neither is present (or when both are sparse), the chunk
carries effectively no searchable content for the figure and
retrieval misses it entirely. Users who configured a local VLM
(reporter's case: Gemma-4-31B) had to post-process MinerU's
`tmp/*.json` themselves.
The deepdoc parser already solves this via
[`VisionFigureParser`](deepdoc/parser/figure_parser.py): when the
tenant has an `IMAGE2TEXT` model configured, each figure gets a
semantic description merged into its chunk. This PR brings the same
behavior to MinerU.
## What changed
### `deepdoc/parser/mineru_parser.py`
- **New method `_enhance_images_with_vlm(outputs, vision_model,
callback=None)`** —
collects every `IMAGE` block with a readable `img_path`, runs
`rag.app.picture.vision_llm_chunk` in a 10-worker
`ThreadPoolExecutor` using the existing
`vision_llm_figure_describe_prompt`, and writes the result back as
`vlm_description`. Per-image failures are logged and skipped — they
never abort the run.
- **`_transfer_to_sections` (IMAGE branch)** — folds
`vlm_description` into the section text alongside caption +
footnote, so the description becomes part of the chunk and is
searchable / retrievable.
- **`parse_pdf`** — after `_read_output`, calls
`_enhance_images_with_vlm(outputs, vision_model, callback=callback)`
when a `vision_model` kwarg is supplied. Wrapped in `try / except`
so a VLM outage cannot break parsing.
### `rag/app/naive.py` (`by_mineru`)
After successfully resolving the MinerU OCR parser, also resolves the
tenant's default `LLMType.IMAGE2TEXT` model via
`get_tenant_default_model_by_type`, wraps it in an `LLMBundle`, and
injects it as `kwargs["vision_model"]` before delegating to
`parse_pdf`.
## Behavior
| Tenant config | Behavior |
|---|---|
| `IMAGE2TEXT` model configured | MinerU image chunks contain `caption +
footnote + VLM description`. Retrieval against figures now actually
works. |
| No `IMAGE2TEXT` model configured | Exact same output as today (caption
+ footnote only). Lookup fails silently with an info log; no error, no
regression. |
| VLM call fails for a single image | That image silently falls back to
caption + footnote; other images proceed. |
| Caller already passes `vision_model` in kwargs | We don't override it
— `if "vision_model" not in kwargs` guards the lookup. |
## Files
- `deepdoc/parser/mineru_parser.py` (+56)
- `rag/app/naive.py` (+13)
## What problem does this PR solve?
Closes#12017.
TTS output is deterministic for a given `(model, text)` pair, so
re-running the same text through the same TTS model produces the same
bytes — yet `Canvas.tts` and `dialog_service.tts` re-synthesized on
every request. That's slow and wastes provider quota whenever the same
assistant response is replayed, shared across users, or repeated within
a session.
### Change
New helper `rag/utils/tts_cache.py` with `synthesize_with_cache(tts_mdl,
cleaned_text)`:
- **Key:** `tts:cache:{model_id}:{sha256(text)}` — separate namespace
per model, identical cleaned text reuses a single entry across both call
sites.
- **Value:** the hex-encoded audio blob both call sites already
returned. No format change for downstream consumers.
- **TTL:** 7 days by default, configurable via
`RAGFLOW_TTS_CACHE_TTL_SECONDS`.
- **Failure modes:** a Redis hiccup falls back to direct synthesis; a
failed synthesis still returns `None` (existing contract preserved).
[`Canvas.tts`](https://github.com/infiniflow/ragflow/blob/main/agent/canvas.py#L683-L724)
and
[`dialog_service.tts`](https://github.com/infiniflow/ragflow/blob/main/api/db/services/dialog_service.py#L1367-L1380)
now route through the helper; the per-file bytes-accumulation/hex-encode
loop has been removed in favor of one shared implementation.
## Type of change
- [x] New Feature (non-breaking change which adds functionality)
## Test plan
- [ ] **Cache hit, chat path:** Configure a dialog with TTS enabled, ask
the same question twice with `stream=false`. Verify the second response
returns the same `audio_binary` and that the second invocation doesn't
hit the TTS provider (e.g., observe provider-side logs / usage counters;
check no `LLMBundle.tts can't update token usage` log line on the second
run).
- [ ] **Cache hit, agent path:** Same exercise via a Conversational
Agent that includes a Message component playing back the answer.
- [ ] **Cache isolation per model:** Switch tenant's `tts_id` between
two models, run the same text against each — confirm the second model's
first synthesis still happens (no cross-model hits).
- [ ] **TTL override:** Set `RAGFLOW_TTS_CACHE_TTL_SECONDS=120`, confirm
the entry expires after 2 minutes.
- [ ] **Redis unavailable:** Stop Redis (or break the connection).
Verify the TTS endpoint still works — synthesis falls back to direct
calls, with a `TTS cache lookup failed` / `TTS cache store failed`
warning logged.
- [ ] **Failure path:** Configure a TTS model with an invalid API key,
ensure the response still returns successfully with `audio_binary=None`
(no regression vs. current behavior).
### What problem does this PR solve?
This PR improves the connector dashboard task management experience and
adds better visibility into connector execution logs.
### Overview:
#### Before
<img width="700" alt="image"
src="https://github.com/user-attachments/assets/e4a8ed6f-2e18-4f0f-8528-41a514550052"
/>
#### Now:
<img width="700" alt="Screenshot from 2026-05-18 16-31-30"
src="https://github.com/user-attachments/assets/d4ca193b-847a-49ae-9e4f-5fbca60ea627"
/>
### 1. Add a new logging page to the connector dashboard
A new logging page has been added so users can view connector task
execution logs directly from the connector dashboard.
### 2. Merge the Resume button into Confirm
The separate **Resume** button has been removed. The **Confirm** button
now represents different actions depending on the current task state:
- **Save**: Save form changes and reschedule tasks.
- **Stop**: Cancel currently scheduled or running tasks.
- **Resume**: Create new scheduled tasks after the previous tasks have
been stopped.
- **Start**: Start tasks when no task has been started yet.
### 3. Separate syncing and pruning tasks
Connector tasks are now separated into **syncing** and **pruning**.
Pruning is controlled by the **Sync deleted files** option:
- When **Sync deleted files** is disabled, only syncing tasks are shown.
- When **Sync deleted files** is enabled, both syncing and pruning tasks
are shown.
**Now: Sync deleted files disabled**
<img width="700" alt="Sync deleted files disabled"
src="https://github.com/user-attachments/assets/dbd9232e-614a-407f-a0b1-c109e5fa567d"
/>
**Now: Sync deleted files enabled**
<img width="700" alt="Sync deleted files enabled"
src="https://github.com/user-attachments/assets/1f527f48-ccb3-4ee8-97ca-086891489296"
/>
### 4. Update logs in backend
<img width="700" alt="image"
src="https://github.com/user-attachments/assets/10a95a3f-98c1-4e67-8afa-ddf6cda5b0b2"
/>
### 5. Remove connector resume API
- Removed: `POST /v1/connectors/<connector_id>/resume`
- Replaced by: `PATCH /v1/connectors/<connector_id>`
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
1. expose batch_chunk_token_size for configuration
2. retrieve chunks when build subgraph for the doc, not retreive all
docs chunks at the begining
3. get all chunks for a document, used to be hard coded 10000
4. delete not used method run_graphrag
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
- [x] Refactoring
Follow on: #14617
## Summary
- Stop pulling chunk vectors (`q_*_vec`) back from Elasticsearch in the
main retrieval path. ES already knows them; shipping them was pure
bandwidth/memory overhead.
- Recover the per-chunk cosine similarity via a second KNN-only ES call
filtered by the candidate chunk ids. The new `_score` is merged with
locally computed term similarity using the user-configured
`vector_similarity_weight`.
- Lazily fetch the chunk embedding only for the chunks
`insert_citations` actually needs.
## Details
**`rag/nlp/search.py`**
- `Dealer.search`: no longer appends `q_*_vec` to the ES select list.
OceanBase still gets it (its rerank path is unchanged).
- New `Dealer._knn_scores(sres, idx_names, kb_ids)`: a `MatchDenseExpr`
over the cached query vector filtered by `id IN sres.ids`, returning
`{chunk_id: cosine_score}` via ES `_score`.
- New `Dealer.rerank_with_knn(...)`: term similarity from
`qryr.token_similarity` plus the ES-supplied KNN score, combined with
`tkweight`/`vtweight` and the existing rank-feature bonus.
- New `Dealer.fetch_chunk_vectors(chunk_ids, tenant_ids, kb_ids, dim)`:
on-demand vector fetch for citation use.
- `Dealer.retrieval` routes Infinity → unchanged, OceanBase → existing
local `rerank`, ES → new KNN-score path.
**`common/doc_store/es_conn_base.py`**
- New `get_scores(res)` helper returning `{_id: _score}` directly from
hit headers (ES doesn't surface `_score` through `get_fields`).
**`api/db/services/dialog_service.py`**
- New top-level `_hydrate_chunk_vectors(...)` helper. On ES it
back-fills `ck["vector"]` from `fetch_chunk_vectors` right before
`insert_citations`. No-op on Infinity / OB (their chunks already carry
vectors).
- Both `decorate_answer` closures became `async` and are `await`-ed at
all call sites in `async_chat` and `async_ask`.
## Backend behavior
| Backend | Returns chunk vec in main search | Sim source | Vectors for
citations |
|---|---|---|---|
| ES | No | second KNN call (`_score`) merged with term sim | fetched on
demand |
| Infinity | No (unchanged) | normalized `_score` | already on chunks |
| OceanBase | Yes (kept) | local hybrid rerank | already on chunks |
## Test plan
## RAG Optimization Description
Optimize the core `BaseTitleChunker` in
`rag/flow/chunker/title_chunker/common.py` to improve RAG document
chunking quality and retrieval accuracy.
## Key Changes
1. **Format-branched text processing**: Preserve original whitespace &
indentation for Markdown/HTML payloads to maintain document semantics
and chunk fidelity; only perform full whitespace cleaning on plain text
content.
2. **Empty chunk filtering**: Thoroughly filter invalid pure-blank lines
to reduce noisy data in vector database.
3. **Code deduplication**: Unified markdown/text/html payload extraction
logic, removed redundant repeated code blocks.
4. **None serialization fix**: Avoid converting `None` value into
literal `"None"` string in chunk text fields.
5. **Production logging**: Added input/output line count logging for
filter logic, observable in online environment.
6. **100% backward compatible**: No changes to chunking hierarchy rules,
output format and all existing workflows.
## RAG Business Value
- Preserves document format fidelity for structured Markdown/HTML files
- Reduces invalid noisy chunks → improves RAG retrieval precision
- Cleans plain text data → optimizes vector embedding quality
- Improves code maintainability with no breaking changes
- Provides observable logging for chunk filtering behavior
## Compatibility
- ✅ No API changes
- ✅ No chunk logic modifications
- ✅ All document parsing/chunking workflows unaffected
- ✅ All pre-checks passed, no code conflicts
### Type of change
- [x] Refactoring
- [x] Performance Improvement
天海蒼灆