Files
ragflow/sdk/python/ragflow_sdk/modules/session.py

198 lines
7.1 KiB
Python
Raw Normal View History

#
# Copyright 2025 The InfiniFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
import json
feat(sdk): make Begin inputs discoverable on Session.ask (#14842) ### What problem does this PR solve? Closes #14751. The user reported that after adding a variable (e.g. `key1`) to an agent's **Begin** component, the Python SDK gave them no way to pass it: their call `session.ask(question=user_question, stream=False)` had no parameter for `key1`, and the `ask()` signature was just `(question, stream, **kwargs)` with a docstring that only described streaming behavior. The functionality already works — `_ask_agent` does `json_data.update(kwargs)` and the server reads `inputs` from the request body at `agent_api.py:902`. The canonical shape is also in the public API docs (`docs/references/python_api_reference.md:1817-1840`): ```python session.ask( "", stream=False, inputs={"line_var": {"type": "line", "value": "I am line_var"}}, return_trace=True, ) ``` But because `inputs`, `release`, and `return_trace` were hidden behind `**kwargs`, they did not appear in IDE signature help, and the docstring did not mention them. Users had no path from "I added a key in the UI" to "I need to pass `inputs=...` with this exact shape." This PR promotes the three most relevant Begin-related arguments to named parameters and rewrites the docstring with a worked example. ### What this PR changes - `sdk/python/ragflow_sdk/modules/session.py`: - `Session.ask()` signature becomes `ask(question="", stream=False, inputs=None, release=None, return_trace=None, **kwargs)`. - These three new named params are forwarded into the existing `kwargs` dict before dispatch, so the wire format and downstream behavior are unchanged. - Docstring rewritten in numpy style, including the structured `{"type": ..., "value": ...}` shape that the Begin component requires (see `agent/component/begin.py:45-60`). No backend changes. `**kwargs` is preserved for forward compatibility with other body fields (`session_id`, `files`, `user_id`, `custom_header`, …). ### Test plan - [ ] `session.ask(question="hi", stream=False)` — existing call still works - [ ] `session.ask("", stream=False, inputs={"key1": {"type": "line", "value": "v"}})` — Begin component receives `key1 = "v"` - [ ] `session.ask("", stream=True, return_trace=True)` — streaming response includes trace events - [ ] IDE / `help(Session.ask)` now shows `inputs`, `release`, `return_trace` with descriptions ### Type of change - [x] Refactoring - [x] Documentation Update
2026-05-19 01:14:57 -07:00
import logging
from .base import Base
feat(sdk): make Begin inputs discoverable on Session.ask (#14842) ### What problem does this PR solve? Closes #14751. The user reported that after adding a variable (e.g. `key1`) to an agent's **Begin** component, the Python SDK gave them no way to pass it: their call `session.ask(question=user_question, stream=False)` had no parameter for `key1`, and the `ask()` signature was just `(question, stream, **kwargs)` with a docstring that only described streaming behavior. The functionality already works — `_ask_agent` does `json_data.update(kwargs)` and the server reads `inputs` from the request body at `agent_api.py:902`. The canonical shape is also in the public API docs (`docs/references/python_api_reference.md:1817-1840`): ```python session.ask( "", stream=False, inputs={"line_var": {"type": "line", "value": "I am line_var"}}, return_trace=True, ) ``` But because `inputs`, `release`, and `return_trace` were hidden behind `**kwargs`, they did not appear in IDE signature help, and the docstring did not mention them. Users had no path from "I added a key in the UI" to "I need to pass `inputs=...` with this exact shape." This PR promotes the three most relevant Begin-related arguments to named parameters and rewrites the docstring with a worked example. ### What this PR changes - `sdk/python/ragflow_sdk/modules/session.py`: - `Session.ask()` signature becomes `ask(question="", stream=False, inputs=None, release=None, return_trace=None, **kwargs)`. - These three new named params are forwarded into the existing `kwargs` dict before dispatch, so the wire format and downstream behavior are unchanged. - Docstring rewritten in numpy style, including the structured `{"type": ..., "value": ...}` shape that the Begin component requires (see `agent/component/begin.py:45-60`). No backend changes. `**kwargs` is preserved for forward compatibility with other body fields (`session_id`, `files`, `user_id`, `custom_header`, …). ### Test plan - [ ] `session.ask(question="hi", stream=False)` — existing call still works - [ ] `session.ask("", stream=False, inputs={"key1": {"type": "line", "value": "v"}})` — Begin component receives `key1 = "v"` - [ ] `session.ask("", stream=True, return_trace=True)` — streaming response includes trace events - [ ] IDE / `help(Session.ask)` now shows `inputs`, `release`, `return_trace` with descriptions ### Type of change - [x] Refactoring - [x] Documentation Update
2026-05-19 01:14:57 -07:00
logger = logging.getLogger(__name__)
class Session(Base):
def __init__(self, rag, res_dict):
self.id = None
self.name = "New session"
self.messages = [{"role": "assistant", "content": "Hi! I am your assistant, can I help you?"}]
for key, value in res_dict.items():
if key == "chat_id" and value is not None:
self.chat_id = None
self.__session_type = "chat"
if key == "agent_id" and value is not None:
self.agent_id = None
self.__session_type = "agent"
super().__init__(rag, res_dict)
feat(sdk): make Begin inputs discoverable on Session.ask (#14842) ### What problem does this PR solve? Closes #14751. The user reported that after adding a variable (e.g. `key1`) to an agent's **Begin** component, the Python SDK gave them no way to pass it: their call `session.ask(question=user_question, stream=False)` had no parameter for `key1`, and the `ask()` signature was just `(question, stream, **kwargs)` with a docstring that only described streaming behavior. The functionality already works — `_ask_agent` does `json_data.update(kwargs)` and the server reads `inputs` from the request body at `agent_api.py:902`. The canonical shape is also in the public API docs (`docs/references/python_api_reference.md:1817-1840`): ```python session.ask( "", stream=False, inputs={"line_var": {"type": "line", "value": "I am line_var"}}, return_trace=True, ) ``` But because `inputs`, `release`, and `return_trace` were hidden behind `**kwargs`, they did not appear in IDE signature help, and the docstring did not mention them. Users had no path from "I added a key in the UI" to "I need to pass `inputs=...` with this exact shape." This PR promotes the three most relevant Begin-related arguments to named parameters and rewrites the docstring with a worked example. ### What this PR changes - `sdk/python/ragflow_sdk/modules/session.py`: - `Session.ask()` signature becomes `ask(question="", stream=False, inputs=None, release=None, return_trace=None, **kwargs)`. - These three new named params are forwarded into the existing `kwargs` dict before dispatch, so the wire format and downstream behavior are unchanged. - Docstring rewritten in numpy style, including the structured `{"type": ..., "value": ...}` shape that the Begin component requires (see `agent/component/begin.py:45-60`). No backend changes. `**kwargs` is preserved for forward compatibility with other body fields (`session_id`, `files`, `user_id`, `custom_header`, …). ### Test plan - [ ] `session.ask(question="hi", stream=False)` — existing call still works - [ ] `session.ask("", stream=False, inputs={"key1": {"type": "line", "value": "v"}})` — Begin component receives `key1 = "v"` - [ ] `session.ask("", stream=True, return_trace=True)` — streaming response includes trace events - [ ] IDE / `help(Session.ask)` now shows `inputs`, `release`, `return_trace` with descriptions ### Type of change - [x] Refactoring - [x] Documentation Update
2026-05-19 01:14:57 -07:00
def ask(
self,
question="",
stream=False,
inputs=None,
release=None,
return_trace=None,
**kwargs,
):
"""
feat(sdk): make Begin inputs discoverable on Session.ask (#14842) ### What problem does this PR solve? Closes #14751. The user reported that after adding a variable (e.g. `key1`) to an agent's **Begin** component, the Python SDK gave them no way to pass it: their call `session.ask(question=user_question, stream=False)` had no parameter for `key1`, and the `ask()` signature was just `(question, stream, **kwargs)` with a docstring that only described streaming behavior. The functionality already works — `_ask_agent` does `json_data.update(kwargs)` and the server reads `inputs` from the request body at `agent_api.py:902`. The canonical shape is also in the public API docs (`docs/references/python_api_reference.md:1817-1840`): ```python session.ask( "", stream=False, inputs={"line_var": {"type": "line", "value": "I am line_var"}}, return_trace=True, ) ``` But because `inputs`, `release`, and `return_trace` were hidden behind `**kwargs`, they did not appear in IDE signature help, and the docstring did not mention them. Users had no path from "I added a key in the UI" to "I need to pass `inputs=...` with this exact shape." This PR promotes the three most relevant Begin-related arguments to named parameters and rewrites the docstring with a worked example. ### What this PR changes - `sdk/python/ragflow_sdk/modules/session.py`: - `Session.ask()` signature becomes `ask(question="", stream=False, inputs=None, release=None, return_trace=None, **kwargs)`. - These three new named params are forwarded into the existing `kwargs` dict before dispatch, so the wire format and downstream behavior are unchanged. - Docstring rewritten in numpy style, including the structured `{"type": ..., "value": ...}` shape that the Begin component requires (see `agent/component/begin.py:45-60`). No backend changes. `**kwargs` is preserved for forward compatibility with other body fields (`session_id`, `files`, `user_id`, `custom_header`, …). ### Test plan - [ ] `session.ask(question="hi", stream=False)` — existing call still works - [ ] `session.ask("", stream=False, inputs={"key1": {"type": "line", "value": "v"}})` — Begin component receives `key1 = "v"` - [ ] `session.ask("", stream=True, return_trace=True)` — streaming response includes trace events - [ ] IDE / `help(Session.ask)` now shows `inputs`, `release`, `return_trace` with descriptions ### Type of change - [x] Refactoring - [x] Documentation Update
2026-05-19 01:14:57 -07:00
Ask a question to the session.
Parameters
----------
question : str
The user's question. May be empty when the agent is driven solely by
Begin component inputs.
stream : bool
If ``True``, yields ``Message`` objects as they arrive (SSE streaming).
If ``False``, yields a single ``Message`` with the final answer.
inputs : dict, optional
Values for variables declared on the agent's **Begin** component. Each
value must be a dict containing at least a ``"value"`` key, and may
include ``"type"``. Example::
session.ask(
"",
stream=False,
inputs={"key1": {"type": "line", "value": "hello"}},
)
Only meaningful for agent sessions; ignored for chat sessions.
release : bool, optional
If ``True``, run against the latest published agent version instead of
the editable draft. Only meaningful for agent sessions.
return_trace : bool, optional
If ``True``, include execution trace information in the response.
Only meaningful for agent sessions.
**kwargs
Additional fields forwarded verbatim to the completion endpoint
(e.g. ``session_id``, ``files``, ``user_id``, ``custom_header``).
See the HTTP API reference for the full list.
"""
feat(sdk): make Begin inputs discoverable on Session.ask (#14842) ### What problem does this PR solve? Closes #14751. The user reported that after adding a variable (e.g. `key1`) to an agent's **Begin** component, the Python SDK gave them no way to pass it: their call `session.ask(question=user_question, stream=False)` had no parameter for `key1`, and the `ask()` signature was just `(question, stream, **kwargs)` with a docstring that only described streaming behavior. The functionality already works — `_ask_agent` does `json_data.update(kwargs)` and the server reads `inputs` from the request body at `agent_api.py:902`. The canonical shape is also in the public API docs (`docs/references/python_api_reference.md:1817-1840`): ```python session.ask( "", stream=False, inputs={"line_var": {"type": "line", "value": "I am line_var"}}, return_trace=True, ) ``` But because `inputs`, `release`, and `return_trace` were hidden behind `**kwargs`, they did not appear in IDE signature help, and the docstring did not mention them. Users had no path from "I added a key in the UI" to "I need to pass `inputs=...` with this exact shape." This PR promotes the three most relevant Begin-related arguments to named parameters and rewrites the docstring with a worked example. ### What this PR changes - `sdk/python/ragflow_sdk/modules/session.py`: - `Session.ask()` signature becomes `ask(question="", stream=False, inputs=None, release=None, return_trace=None, **kwargs)`. - These three new named params are forwarded into the existing `kwargs` dict before dispatch, so the wire format and downstream behavior are unchanged. - Docstring rewritten in numpy style, including the structured `{"type": ..., "value": ...}` shape that the Begin component requires (see `agent/component/begin.py:45-60`). No backend changes. `**kwargs` is preserved for forward compatibility with other body fields (`session_id`, `files`, `user_id`, `custom_header`, …). ### Test plan - [ ] `session.ask(question="hi", stream=False)` — existing call still works - [ ] `session.ask("", stream=False, inputs={"key1": {"type": "line", "value": "v"}})` — Begin component receives `key1 = "v"` - [ ] `session.ask("", stream=True, return_trace=True)` — streaming response includes trace events - [ ] IDE / `help(Session.ask)` now shows `inputs`, `release`, `return_trace` with descriptions ### Type of change - [x] Refactoring - [x] Documentation Update
2026-05-19 01:14:57 -07:00
if inputs is not None:
kwargs["inputs"] = inputs
if release is not None:
kwargs["release"] = release
if return_trace is not None:
kwargs["return_trace"] = return_trace
if inputs is not None or release is not None or return_trace is not None:
logger.debug(
"Session.ask explicit-params session_type=%s session_id=%s "
"input_keys=%s release=%s return_trace=%s",
self.__session_type,
getattr(self, "id", None),
list(inputs.keys()) if isinstance(inputs, dict) else None,
release,
return_trace,
)
if self.__session_type == "agent":
res = self._ask_agent(question, stream, **kwargs)
elif self.__session_type == "chat":
res = self._ask_chat(question, stream, **kwargs)
else:
raise Exception(f"Unknown session type: {self.__session_type}")
if stream:
for line in res.iter_lines(decode_unicode=True):
if not line:
continue # Skip empty lines
line = line.strip()
if line.startswith("data:"):
content = line[len("data:"):].strip()
if content == "[DONE]":
break # End of stream
else:
content = line
try:
json_data = json.loads(content)
except json.JSONDecodeError:
continue # Skip lines that are not valid JSON
event = json_data.get("event",None)
if event and event != "message":
continue
if (
(self.__session_type == "agent" and event == "message_end")
or (self.__session_type == "chat" and json_data.get("data") is True)
):
return
if self.__session_type == "agent":
yield self._structure_answer(json_data)
else:
yield self._structure_answer(json_data["data"])
else:
try:
json_data = res.json()
except ValueError:
raise Exception(f"Invalid response {res}")
yield self._structure_answer(json_data["data"])
def _structure_answer(self, json_data):
answer = ""
if self.__session_type == "agent":
answer = json_data["data"]["content"]
elif self.__session_type == "chat":
answer = json_data["answer"]
reference = json_data.get("reference", {})
temp_dict = {
"content": answer,
"role": "assistant"
}
if reference and "chunks" in reference:
chunks = reference["chunks"]
temp_dict["reference"] = chunks
message = Message(self.rag, temp_dict)
return message
def _ask_chat(self, question: str, stream: bool, **kwargs):
json_data = {"question": question, "stream": stream, "session_id": self.id}
json_data.update(kwargs)
res = self.post(f"/chats/{self.chat_id}/completions",
json_data, stream=stream)
return res
def _ask_agent(self, question: str, stream: bool, **kwargs):
Feat: Agent api (#14157) ### What problem does this PR solve? 1. **List agents** **Prev API**: - `/v1/canvas/list GET` - `/api/v1/agents GET` **Current API**: `/api/v2/agents GET` 2. **Get canvas template** **Prev API**: `/v1/canvas/templates GET` **Current API**: `/api/v2/agents/templates GET` 3. **Delete an agent** **Prev API**: - `/v1/canvas/rm POST` - `/api/v1/agents/<agent_id> DELETE` **Current API**: `/api/v2/agents/<agent_id> DELETE` 4. **Update an agent** **Prev API**: - `/api/v1/agents/<agent_id> PUT` - `/v1/canvas/setting POST ` **Current API**: `/api/v2/agents/<agent_id> PATCH` 5. **Create an agent** **Prev API**: - `/v1/canvas/set POST` - `/api/v1/agents POST` **Current API**: `/api/v2/agents POST` 6. **Get an agent** **Prev API**: - `/v1/canvas/get/<canvas_id> GET ` **Current API**: `/api/v2/agents/<agent_id> GET` 7. **Reset an agent** **Prev API**: - `/v1/canvas/reset POST` **Current API**: `/api/v2/agents/<agent_id>/reset POST` 8. **Upload a file to an agent** **Prev API**: - `/v1/canvas/upload/<canvas_id> POST` **Current API**: `/api/v2/agents/<agent_id>/upload POST` 9. **Input form** **Prev API**: - `/v1/canvas/input_form GET` **Current API**: `/api/v2/agents/<agent_id>/components/<component_id>/input-form GET` 10. **Debug an agent** **Prev API**: - `/v1/canvas/debug POST` **Current API**: `/api/v2/agents/<agent_id>/components/<component_id>/debug POST` 11. **Trace an agent** **Prev API**: - `/v1/canvas/trace GET` **Current API**: `/api/v2/agents/<agent_id>/logs/<message_id> GET` 12. **Get an agent version list** **Prev API**: - `/v1/canvas/getlistversion/<canvas_id>` **Current API**: `/api/v2/agents/<agent_id>/versions GET` 13. **Get a version of agent** **Prev API**: - `/v1/canvas/getversion/<version_id>` **Current API**: `/api/v2/agents/<agent_id>/versions/<version_id> GET` 14. **Test db connection** **Prev API**: - `/v1/canvas/test_db_connect POST` **Current API**: `/api/v2/agents/test_db_connection` 15. **Rerun the agent** **Prev API**: - `/v1/canvas/rerun POST` **Current API**: `/api/v2/agents/rerun POST` 16. **Get prompts** **Prev API**: - `/v1/canvas/prompts GET` **Current API**: `/api/v2/agents/prompts GET` ### Type of change - [x] New Feature (non-breaking change which adds functionality) --------- Co-authored-by: chanx <1243304602@qq.com>
2026-04-24 10:02:22 +08:00
json_data = {
"agent_id": self.agent_id,
"query": question,
"stream": stream,
"session_id": self.id,
"openai-compatible": False,
}
json_data.update(kwargs)
res = self.post("/agents/chat/completions", json_data, stream=stream)
return res
def update(self, update_message):
res = self.patch(f"/chats/{self.chat_id}/sessions/{self.id}",
update_message)
res = res.json()
if res.get("code") != 0:
raise Exception(res.get("message"))
class Message(Base):
def __init__(self, rag, res_dict):
self.content = "Hi! I am your assistant, can I help you?"
self.reference = None
self.role = "assistant"
self.prompt = None
self.id = None
super().__init__(rag, res_dict)