marcolardera
feat: add anthropic
stuck at error invalid API Key. To reproduce:
export CHAT_API_CLI=xxx <anthropic key>
chatgpt-cli --model claude-3-5-sonnet-20240620 --supplier anthropic
Warning: current state works for Anthropic, but somehow breaks the openai integration.
The mix between biz logic / config logic and networking logic makes scaling out this CLI kind of hard. Would recommend rewrite the logic before scaling out to other integrations.
https://github.com/tadeasf/chatgpt-cli/tree/styling @TiansuYu pretty much rewrote it... using litellm sdk https://docs.litellm.ai/docs/ for scalling.. so now you should be able to use pretty much all the models
- i have bug in /c -> struggling with tagging the blocks as copy
- included saving history in markdown or json option.. i prefer markdown as I am dumping it to my obsidian vault and the grepping through ti - pretty neat
- included budgetmanager from litellm as well, not production though so you are limited per session but i will fine tune it eventually as well
- included /e -> opens the markdown output in your set $EDITOR so you can copy what you need in case of big output the prompt library is unable to handle
- refactored it etc will need general cleanup it's not really 100% ready yet but all the core functions work and are built on top of litellm
after i do general cleanup and get time to update readme, setup.cfg, configs etc i will push "beta" or whatnot
before i do that I am thinking about implementing a few general budget friendly options e.g. using embeddings models to map up your repo/project/long scripts before sending it to the llm for output and shit like that
i wrote it mainly for myself but i am very much willing to incorporate any hints and tips you throw at me
oh, i also included prompt completer for models (if you pick a wrong one which is validated through litellm) and then path completer cuz nobody likes to have typos in paths haha
I streamlined the UI, updated readme, fixed a few bugs, included session-wide budget management in ~/.config/chatgpt-cli/cost_usage.json which you can set easily to monthly/daily stuff. tbh, i haven't expected to put so much time into it, but i wanted to try out tiangolo's AWESOME typer cli framework and here we are haha
https://github.com/user-attachments/assets/c3ba47f6-3c0a-4aa1-9413-a9a9dea275f2
Hey @tadeasf That looks awesome! Cannot wait for this to get out!
feel free to go here follow the instructions in the rewritten readme and you should be able to install it via pip or you can build it via pdm
the big llm part is practically finished. now i am working on ollama support... so I think that maybe i will do a PR here and find out whether @marcolardera is okay with me raping his initial one-script based solution like this or whether I should kill the fork and spin up a whole new project alltogether :)
Just tested against openai, look all good. Though I find adding this to the end of every chat, is a bit verbose:
ChatGPT CLI
Provider: openai
Model: gpt-4o
───────────────────────────────────
[Tokens: 1076] [Cost: $0.029787]
And what I am missing from the original code is a schema for the config file. I wish there could be a dataclass or something for me to quickly understand how to configure this thing.
Also there is no CLI override options, which I think could be improved.
Let me know if you want a lending hand, we could work this together.
Which branch you tested? Main is behind do styling. Lending hand 100% appreciated I was trying to push it but doing too much open source at the moment so second sets of eyes are very lovely
@marcolardera What do you think of this overhaul initiative? Would you consider this to be merged into repo? Or you prefer us working on something separately.