cloud-providers
local LLMs
kuzco was originally designed to work with local LLMs through “ollamar.” These models are run using Ollama, which supports both “ollamar” and “ellmer” backends. After installing Ollama on your system, you need to pull a model to use locally.
- install Ollama
- pull a model (preferably a vision model) such as qwen2.5vl
- install kuzco
- run your AI enabled computer vision tasks
Ollama has served as the default provider for some time, check out getting-started or the README for local examples. But kuzco can now also utilize cloud-based providers.
cloud LLMs
In recent months, kuzco has attracted significant interest from organizations like R-Consortium and Posit, as well as from individuals curating lists of LLM and AI projects for R and open-source communities. Many users have highlighted the need for supporting non-local LLMs, whether due to hardware limitations or a desire to leverage their API account with Mistral, Google, OpenAI, etc. Given this feedback, adding non-local LLM functionality to kuzco is a logical next step.
cloud example
Note that provider = "ollama" is the default, so this change does not interfere with legacy processes. However, when the “ellmer” backend is selected (which is also the default), you have the option to change the provider as needed.
Note below showcases the use of the Mistral provider.
Since cloud providers require an API key, the API key can be set like so:
# via base R:
Sys.setenv(MISTRAL_API_KEY = "the_api_key_via_the_provider")
# or usethis:
usethis::edit_r_environ()This is the same exact environment variable that you would use via ellmer.
If you’re unsure how to name your API key, refer to the documentation for each provider in the ellmer documentation for each chat provider.
Once you’ve set the environment variable, running computer vision tasks becomes straightforward:
kuzco::llm_image_classification(provider = "mistral", llm_model = "pixtral-12b", image = test_img)cloud provider dictionary
For your reference, here is a complete list of “providers” available in kuzco, along with their corresponding ellmer “chat” functions:
| provider | ellmer..function |
|---|---|
| ollama | ellmer::chat_ollama |
| anthropic | ellmer::chat_anthropic |
| aws_bedrock | ellmer::chat_aws_bedrock |
| azure | ellmer::chat_azure |
| azure_openai | ellmer::chat_azure_openai |
| bedrock | ellmer::chat_bedrock |
| claude | ellmer::chat_claude |
| cloudfare | ellmer::chat_cloudflare |
| cortex | ellmer::chat_cortex |
| cortex_analyst | ellmer::chat_cortex_analyst |
| databricks | ellmer::chat_databricks |
| deepseek | ellmer::chat_deepseek |
| gemini | ellmer::chat_gemini |
| github | ellmer::chat_github |
| google_gemini | ellmer::chat_google_gemini |
| google_vertex | ellmer::chat_google_vertex |
| groq | ellmer::chat_groq |
| huggingface | ellmer::chat_huggingface |
| mistral | ellmer::chat_mistral |
| openai | ellmer::chat_openai |
| openrouter | ellmer::chat_openrouter |
| perplexity | ellmer::chat_perplexity |
| portkey | ellmer::chat_portkey |
| snowflake | ellmer::chat_snowflake |
| vllm | ellmer::chat_vllm |