ChatWithTree gramplet addition #762
Conversation
|
You only need to include the The |
|
Looking forward to testing thank you for this addon 👍 @Nick-Hall Attempted to test on Windows with the GrampsAIO (All in one) but the GrampsAIO install looks to be missing the required files (see error message below). ModuleNotFoundError: No module named 'ctypes.wintypes' fyi on windows I had to select the option below first before getting the error, can you make the gramplet more graceful so it displays the gramplet with a message showing that python module litellm needs to be installed 
|
|
Attempted to test on MacOS, which is also a bundled gramps distro. The litellm dependency has about 45 other dependencies. Even after installing all of these into the gramps bundle, still had issues with pydantic_core wheel. Will attempt again when linux testing is successful (by others). Open issues:
|
MelleKoning
force-pushed
the
myaddon60
branch
3 times, most recently
from
245242e to
7cfc554
Compare
|
Thanks for the swift and extensive comments and the test on windows. I have adjusted the commit by:
For the required llm module, the "litellm" is part of the ChatWithTree.gpr.py registration file - in my testing the module is installed and working. Maybe litellm is dependend on other modules that have to be accessible by Python? Locally, I installed and ran the plugin with the gramps version 6.0.4, the development version. Locally, I have python 3.12.3 (python3 --version) As I work in a python virtual environment, this gives me the oppertunity to generate a requirements.txt. These are the modules that can be seen from that requirements.txt file: This does not show as much as I thought, so I used another method with I now get a huge list, that also includes litellm, this file requirements.txt is in the main addons-source folder, so I think it includes ALL packages needed by all the addons and gramps: There is also a requirements.txt generated in the /ChatWithTree subfolder only: I do not understand why the litellm version here is different from the one above (1.76.1 vs 1.75.7), but it has far less dependencies. About Windows: maybe the installed python3 or litellm version, or one of its dependencies, is not matching with the above? |
| try: | ||
| from typing import Dict, Any, List, Optional, Tuple, Pattern, Iterator | ||
| import os | ||
| import json | ||
| import sys | ||
| import time | ||
| import re | ||
| import inspect | ||
| import litellm | ||
| except ImportError as e: | ||
| LOG.warning(e) | ||
| raise Exception("ChatWithTree requires litellm") |
There was a problem hiding this comment.
Putting a try/except around such large blocks is not a good practice, and can hide issues.
There was a problem hiding this comment.
I tend to agree - most of the imports seem to be default modules (os, json, sys, time), but am not entirely sure about inspect, re
As litellm is a requirement, just as the other imports, what is the best alternative suggested:
- Put a try..except around each import
- Only have the try..except around the litellm module (as that is what we want to test)
- Or maybe - don't have the try..except at all
There was a problem hiding this comment.
Everything but "litellm" here is included with python.
| def _llm_loop(self, seed: int) -> Iterator[Tuple[YieldType, str]]: | ||
| # Tool-calling loop | ||
| final_response = "I was unable to find the desired information." | ||
| limit_loop = 6 |
There was a problem hiding this comment.
Not sure that will be enough. Maybe make this a parameter, or config setting?
There was a problem hiding this comment.
I found two different issues with a higher number
- Original number 10 could cause "same questions asked to the llm" especially with very dumb llms (local ollama llms that are small are not that smart, although they do work with litellm). It was kind of wasting time without making any progress in the roundtrips
- Rate limits of remote llms. When doing 10 quick roundtrips to the remote llms you often quickly hit a rate limit, while if we limit it to about 5 calls, and then let the user type a new question, rarely hitting the rate limit of the remote cloud llms
However: Limiting from 10 originally back to max 6 seems good enough, especially with models that are able to instruct to do multiple of the local tool calls at once, that is, the number of tool calls on the local system is not limited - this is important.
As an example - here is a question whereby in the chat there were already plenty people retrieved, and then I asked can you show me all birth and death dates for all these family members?
This is the tool-call balloon in yellow:
So - the 6 is about the number of roundtrips to the (remote) language model but not limiting the amount of tool calls to the local database - which can be pretty huge as seen above. Does that make sense?
Nevertheless, yes:
- I do agree that making this "6" configurable is a good enhancement. I could probably best use the existing "/commands" pattern for that and add this configurable option to the /help page. Would love to do that in an upcoming MR.
| person_obj = self.db.get_person_from_handle(person_handle) | ||
| obj = self.sa.mother(person_obj) | ||
| data = dict(self.db.get_raw_person_data(obj.handle)) | ||
| return data |
There was a problem hiding this comment.
The SimpleAccess is easy, but there are better ways to do what you want more directly. There may be more than one mother, for example.
|
|
||
| if family_handle_list: | ||
| family_id = family_handle_list[0] | ||
| family = self.db.get_family_from_handle(family_id) |
There was a problem hiding this comment.
Using the raw functions will be much faster for such uses.
| #search_pattern = re.compile(re.escape(search_string), re.IGNORECASE) | ||
| search_pattern = self.create_search_pattern(search_string) | ||
|
|
||
| for person_obj in self.sa.all_people(): |
There was a problem hiding this comment.
For personal use asking for all people with the name "John" it's quick enough - The performance comments probably highly depend on the size of the database? For now only have used the SimpleAccess database entry point.
Enhancing that to use other patterns to ensure that Gramps users with many more data in the database can also use this tool is definately on the to do list.
What area's to look for within gramps to get a better understanding on how to use raw data access as opposed to the SimpleAccess?
There was a problem hiding this comment.
I'm going to work on a set of functions tonight/tomorrow.
| try: | ||
| from ChatWithTreeBot import ChatBot | ||
| except ImportError as e: | ||
| LOG.warning(e) | ||
| raise ImportError("Failed to import ChatBot from chatbot module: " + str(e)) | ||
|
|
There was a problem hiding this comment.
This try/except doesn't do much more than without it.
|
Looks indeed to be a missing library ctypes.wintypes, a standard .py file in the Windows AIO installer. The Windows GrampsAIO64-6.0.4 installer's Python environment is missing this standard library in it's library.zip on your machine. I would not recommend trying to add the python file to that library.zip yourself but it could be a workaround. This goes beyond this particular addon. If there is an existing script that creates the windows installer we can inspect why this library is missing. |
|
I installed litellm (and its 45 dependencies) to the ARM MacOS bundle of my 6.0.4 app so I coud test this addon. When I ran ChatWithTree addon, I got an error below. Can you provide any insights on this. BTW, I had a similar issue [not valid for use in process: mapping process and mapped file (non-platform) have different Team IDs] when I installed pygraphviz and tried to run the NetworkChart addon. 2025-09-05 10:22:34.309: WARNING: ChatWithTreeBot.py: line 33: dlopen(/Volumes/Storage/Applications/Gramps6.0.4.app/Contents/Resources/lib/python3.13/site-packages/pydantic_core/_pydantic_core.cpython-313-darwin.so, 0x0002): tried: '/Volumes/Storage/Applications/Gramps6.0.4.app/Contents/Resources/lib/python3.13/site-packages/pydantic_core/_pydantic_core.cpython-313-darwin.so' (code signature in <2A10A0C5-9ABF-35D2-B314-198B0E3F4A16> '/Volumes/Storage/Applications/Gramps6.0.4.app/Contents/Resources/lib/python3.13/site-packages/pydantic_core/_pydantic_core.cpython-313-darwin.so' not valid for use in process: mapping process and mapped file (non-platform) have different Team IDs), '/System/Volumes/Preboot/Cryptexes/OS/Volumes/Storage/Applications/Gramps6.0.4.app/Contents/Resources/lib/python3.13/site-packages/pydantic_core/_pydantic_core.cpython-313-darwin.so' (no such file), '/Volumes/Storage/Applications/Gramps6.0.4.app/Contents/Resources/lib/python3.13/site-packages/pydantic_core/_pydantic_core.cpython-313-darwin.so' (code signature in <2A10A0C5-9ABF-35D2-B314-198B0E3F4A16> '/Volumes/Storage/Applications/Gramps6.0.4.app/Contents/Resources/lib/python3.13/site-packages/pydantic_core/_pydantic_core.cpython-313-darwin.so' not valid for use in process: mapping process and mapped file (non-platform) have different Team IDs) During handling of the above exception, another exception occurred: Traceback (most recent call last): |
|
@GaryGriffin: It looks like it added libraries to the bundle that somebody else signed. I can disable that check with an entitlement. The description of that entitlement says that it turns on extra Gatekeeper checks so I'll have to make a bundle to test-notarize and then for you to try out. |
...
Although I don't have a MAC, this indicates a trust issue from the library you added to gramps with the signature of the installed gramps 6.0.4. When gramps project tries to load this module pydentic_core.cpython, gramps does not trust it. Apparently you had a similar issue with another addon. I guess that is a good safety measure of a Mac to only allow *.so compiled libraries installed from the same application (Gramps) instead of allowing access to self-installed compiled binaries as downloaded from the web. Maybe it is possible to override this? Removing a quaraintaine flag, ref: Maybe this helps: Goes a bit beyond the MR at hand, but it is clear that the litellm packge is not installed by default with the gramps installation, and also not easily installed from within the addon manager of gramps when gramps is running; now seen both on Mac as previously on Windows by @PQYPLZXHGF |
|
@GaryGriffin: https://sourceforge.net/projects/gramps/files/Testing/Gramps-Arm-6.0.4-NLV.dmg/download NLV stands for no library verification but I gather from the docs that it really means that libraries must still be signed, just not necessarily by me. @Nick-Hall said something yesterday about maybe releasing 6.0.5 this weekend so if you can verify that this works I'll do the same for the release. |
|
@jralls No core file created. |
|
Rats. It seems there's an error in the Apple Docs and disable-library-validation requires a provisioning profile--but what capability is needed isn't documented so I've asked for help. |
|
@Nick-Hall do you know where on windows the site-packages get installed when a plugin/addon wants to install site-packages? (tag @PQYPLZXHGF ) On a Windows 10 machine, manually I was able to add a file Within Gramps the addon installer then goes a bit further until it spits out another error about pip not able to install the addon. Then, manually, we can still install the
I used this manual command, because I saw that from the code The above command nstalls litellm and dependend packages into the So question:
I hope by manually installing the litellm dependency, outside of gramps, installing of this addon will succeed, because the dependency Let me know if the above made any sense, or none at all - otherwise, how can we update the Gramps AIO installer to already include certain dependencies that addons might need, like this |
We install python libraries for addons into the LIB_PATH directory which is called "lib" and located under the user plugins directory. We add this to the python path. Running I'll create a Windows AIO that contains |
@GaryGriffin It turned out I'd made a typo in the entitlements file. I fixed that and made a 6.0.5-NLV for you to try. |
ChatWithTree plugin addition.
As this is my first MR for a Gramps plugin - please advice.
Will add a topic in the gramps development forum with the features:
https://gramps.discourse.group/t/chatwithtree-plugin-gramplet/8251