### What problem does this PR solve?
The OpenAI-compatible chat endpoint
(`/chats_openai/<chat_id>/chat/completions`) was not returning accurate
token
usage in streaming responses. The token counts were either missing or
inaccurate because the underlying LLM API
responses weren't being properly parsed for usage data.
This PR adds proper token counting using tiktoken (cl100k_base encoding)
as a fallback when the LLM API doesn't provide usage data in streaming
chunks. This ensures clients always receive token usage information in
the
response, which is essential for billing and quota management.
**Changes:**
- Add tiktoken-based token counting for streaming responses in
OpenAI-compatible endpoint
- Ensure `usage` field is always populated in the final streaming chunk
- Add unit tests for token usage calculation
Fixes#7850
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Feat: Add a web search button to the chat box on the chat page.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
**Backend**
\rag\nlp\search.py
*Before the fix*
The top_k parameter was not applied to limit the total number of chunks,
and the rerank model also uses the exact whole valid_idx rather than
assigning valid_idx = valid_idx[:top] firstly.
*After the fix*
The top_k limit is applied to the total results before pagination, using
a default value of top = 1024 if top_k is not modified.
session.py
*Before the fix:*
When the frontend calls the retrieval API with `search_id`, the backend
only reads `meta_data_filter` from the saved `search_config`. The
`rerank_id`, `top_k`, `similarity_threshold`, and
`vector_similarity_weight` parameters are only taken from the direct
request body. Since the frontend doesn't pass these parameters
explicitly (it only passes `search_id`), they always fall back to
default values:
- `similarity_threshold` = 0.0
- `vector_similarity_weight` = 0.3
- `top_k` = 1024
- `rerank_id` = "" (no rerank)
This means user settings saved in the Search Settings page have no
effect on actual search results.
*After the fix:*
When a `search_id` is provided, the backend now reads all relevant
configuration from the saved `search_config`, including `rerank_id`,
`top_k`, `similarity_threshold`, and `vector_similarity_weight`. Request
parameters can still override these values if explicitly provided,
allowing flexibility. The rerank model is now properly instantiated
using the configured `rerank_id`, making the rerank feature actually
work.
**Frontend**
\web\src\pages\next-search\search-setting.tsx
*Before the fix*
search-setting.tsx file, the top_k input box is only displayed when
rerank is enabled (wrapped in the rerankModelDisabled condition). If the
rerank switch is turned off, the top_k input field will be hidden, but
the form value will remain unchanged. In other words: - When rerank is
enabled, users can modify top_k (default 1024). - When rerank is
disabled, top_k retains the previous value, but it's not visible on the
interface. Therefore, the backend will always receive the top_k
parameter; it's just that the frontend UI binds this configuration item
to the rerank switch. When rerank is turned off, top_k will not
automatically reset to 1024, but will retain its original value.
*After the fix*
On the contrary, if we switch off the button rerank model, the value
top-k will be reset to 1024. By the way, If we use top-k in an
individual method, rather than put it into the method retrieval, we can
control it separately
Now all methods valid
Using rerank
<img width="2378" height="1565" alt="Screenshot 2026-01-21 190206"
src="https://github.com/user-attachments/assets/fa2b0df0-1334-4ca3-b169-da6c5fd59935"
/>
Not using rerank
<img width="2596" height="1559" alt="Screenshot 2026-01-21 190229"
src="https://github.com/user-attachments/assets/c5a80522-a0e1-40e7-b349-42fe86df3138"
/>
Before fixing they are the same
### Type of change
- Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
API adds audio to text and text to speech functions
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
change:
Enhance delta streaming in chat functions for improved reasoning and
content handling
### Type of change
- [x] Refactoring
## Summary
Fixes#12311
Changes the `reference` field initialization from `[{}]` to `[]` in
session creation.
### Problem
When creating a session via the SDK API, the `reference` field was
incorrectly initialized as `[{}]`. This caused:
- First dialogue round: Empty reference
- Second dialogue round: Reference pointing to first round's data
- Overall misalignment between dialogue rounds and their references
### Solution
Changed the initialization to `[]` (empty list), which:
- Matches the `Conversation` model's expected default
- Ensures references grow correctly one-to-one with assistant responses
- Aligns with the service layer's expectations
### Testing
After applying this fix:
1. Create a session via `POST /api/v1/chats/{conversation_id}/sessions`
2. Send multiple questions via `POST
/api/v1/chats/{conversation_id}/completions`
3. View the conversation on web - references should now align correctly
with each dialogue round
### What problem does this PR solve?
Manage message and use in agent.
Issue #4213
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Chats completions API supports metadata filtering.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Trace information can be returned by the agent completion API.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Refactor metadata filter.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] Refactoring
---------
Co-authored-by: Kevin Hu <kevinhu.sh@gmail.com>
### What problem does this PR solve?
Add DeepseekV3.2 of Tongyi-Qianwen and remove unused code
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
change：
async issue and sensitive logging
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Retrieval metadata filtering adds semi-automatic mode, and users can
manually check the metadata key that participates in LLM to generate
filter conditions.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
Cleanup synchronous functions in chat_model and implement
synchronization for conversation and dialog chats.
### Type of change
- [x] Refactoring
- [x] Performance Improvement
### What problem does this PR solve?
Make RAGFlow more asynchronous 2. #11551, #11579, #11619.
### Type of change
- [x] Refactoring
- [x] Performance Improvement
### What problem does this PR solve?
Make RAGFlow more asynchronous 2. #11551, #11579, #11619.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
- [x] Refactoring
- [x] Performance Improvement
### What problem does this PR solve?
Try to make this more asynchronous. Verified in chat and agent
scenarios, reducing blocking behavior. #11551, #11579.
However, the impact of these changes still requires further
investigation to ensure everything works as expected.
### Type of change
- [x] Refactoring
### What problem does this PR solve?
1. Update RetCode to common.constants
2. Decouple the admin and API modules
### Type of change
- [x] Refactoring
---------
Signed-off-by: Jin Hai <haijin.chn@gmail.com>
### What problem does this PR solve?
1. Rename identifier name
2. Fix some return statement
3. Fix some typos
### Type of change
- [x] Refactoring
Signed-off-by: Jin Hai <haijin.chn@gmail.com>
### What problem does this PR solve?
Add get_uuid, download_img and hash_str2int into misc_utils.py
### Type of change
- [x] Refactoring
---------
Signed-off-by: Jin Hai <haijin.chn@gmail.com>
### What problem does this PR solve?
Revert back to chat.completions.
### Type of change
- [ ] Bug Fix (non-breaking change which fixes an issue)
- [ ] New Feature (non-breaking change which adds functionality)
- [ ] Documentation Update
- [ ] Refactoring
- [ ] Performance Improvement
- [x] Other (please describe):
Revert back to chat.completions.
### What problem does this PR solve?
Fix broken imports
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
---------
Signed-off-by: jinhai <haijin.chn@gmail.com>
### What problem does this PR solve?
Fix: resolve hash collisions by switching to UUID &correct logic for
always-true statements, solved: #10165
Feat: Update GPT api integration, solved: #10204
Feat: Support qianwen-deepresearch, solved: #10163
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] New Feature (non-breaking change which adds functionality)
### What problem does this PR solve?
An exception happens if you give session_id to agent_open_ai completion.
Because session_id is being given as well as **req so it tries to send
session_id twice. But also the logic seemed odd on picking one of
session_id, id, metadata.id. So cleaned it up a little.
See #10111
### Type of change
- [X] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Unify reference format of agent completion and OpenAI-compatible
completion API.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
- [x] Documentation Update
- [x] Refactoring
### What problem does this PR solve?
Revert broken agent completion by #9631.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
This PR fixes a critical bug in the session listing endpoint where the
application crashes with an `AttributeError` when processing chunk data
that contains non-dictionary objects.
**Error before fix:**
```json
{
"code": 100,
"data": null,
"message": "AttributeError(\"'str' object has no attribute 'get'\")"
}
```
**Root cause:**
The code assumes all items in the `chunks` array are dictionary objects
and directly calls the `.get()` method on them. However, in some cases,
the chunks array contains string objects or other non-dictionary types,
causing the application to crash when attempting to call `.get()` on a
string.
**Solution:**
Added type validation to ensure each chunk is a dictionary before
processing. Non-dictionary chunks are safely skipped, preventing the
crash while maintaining functionality for valid chunk data.
This fix improves the robustness of the session listing endpoint and
ensures users can retrieve their conversation sessions without
encountering server errors due to data format inconsistencies.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
When create conversation,the prologue hasn't save in conversation.
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Fix search app AI summary ERROR: 'dict' object has no attribute 'split'.
#9649
### Type of change
- [x] Bug Fix (non-breaking change which fixes an issue)
### What problem does this PR solve?
Resolve#9549 and #9436 , In v0.20.x,Agent completions API changed a
lot,such as without reference and so on
### Type of change
- [x] Refactoring
