<!-- README.md is generated from README.Rmd. Please edit that file -->
# ellmer <a href="https://ellmer.tidyverse.org"><img src="man/figures/logo.png" align="right" height="138" alt="ellmer website" /></a>
<!-- badges: start -->
[](https://lifecycle.r-lib.org/articles/stages.html#experimental)
[](https://github.com/tidyverse/ellmer/actions/workflows/R-CMD-check.yaml)
<!-- badges: end -->
ellmer makes it easy to use large language models (LLM) from R. It
supports a wide variety of LLM providers and implements a rich set of
features including streaming outputs, tool/function calling, structured
data extraction, and more.
ellmer is one of a number of LLM-related packages created by Posit:
- Looking for something similar in python? Check out
[chatlas](https://github.com/posit-dev/chatlas)!
- Want to evaluate your LLMs? Try
[vitals](https://vitals.tidyverse.org).
- Need RAG? Take a look at [ragnar](https://ragnar.tidyverse.org).
- Want to make a beautiful LLM powered chatbot? Consider
[shinychat](https://posit-dev.github.io/shinychat/).
- Working with MCP? Check out
[mcptools](https://posit-dev.github.io/mcptools/).
## Installation
You can install ellmer from CRAN with:
``` r
install.packages("ellmer")
```
## Providers
ellmer supports a wide variety of model providers. Official providers
are actively maintained, with priority support for bug fixes and new
features. Community providers are contributed and maintained by the
community; contributions to improve them are especially welcome.
### Official providers
- Anthropicโs Claude: `chat_anthropic()`.
- AWS Bedrock: `chat_aws_bedrock()`.
- Azure OpenAI: `chat_azure_openai()`.
- Databricks: `chat_databricks()`.
- DeepSeek: `chat_deepseek()`.
- Google Gemini/Vertex AI: `chat_google_gemini()`,
`chat_google_vertex()`.
- Ollama: `chat_ollama()`.
- OpenAI: `chat_openai()`.
- Posit AI: `chat_posit()`.
- Snowflake Cortex: `chat_snowflake()`.
### Community providers
- Cloudflare: `chat_cloudflare()`.
- Groq: `chat_groq()`.
- Hugging Face: `chat_huggingface()`.
- LM Studio: `chat_lmstudio()`.
- Mistral: `chat_mistral()`.
- OpenRouter: `chat_openrouter()`.
- perplexity.ai: `chat_perplexity()`.
- Portkey: `chat_portkey()`.
- VLLM: `chat_vllm()`.
### Provider/model choice
If youโre using ellmer inside an organisation, you may have internal
policies that limit you to models from big cloud providers,
e.g.ย `chat_azure_openai()`, `chat_aws_bedrock()`, `chat_databricks()`,
or `chat_snowflake()`.
If youโre using ellmer for your own exploration, youโll have a lot more
freedom, so we have a few recommendations to help you get started:
- `chat_openai()` or `chat_anthropic()` are good places to start.
`chat_openai()` defaults to **GPT-6.1 Terra**, but you can use
`model = "gpt-6-luna"` for a cheaper, faster model.
`chat_anthropic()` defaults to **Claude Sonnet 5.5**, which we have
found to be particularly good at writing R code.
- `chat_google_gemini()` is a strong model with a free tier (with the
downside that [your data is
used](https://aistudio.google.com/docs/terms#unpaid-services) to
improve the model), making it a great place to start if you donโt want
to spend any money.
- `chat_ollama()`, which uses [Ollama](https://ollama.com), allows you
to run models on your own computer. While the biggest models you can
run locally arenโt as good as the state of the art hosted models, they
donโt share your data and are effectively free.
### Authentication
Authentication works a little differently depending on the provider. A
few popular ones (including OpenAI and Anthropic) require you to obtain
an API key. We recommend you save it in an environment variable rather
than using it directly in your code, and if you deploy an app or report
that uses ellmer to another system, youโll need to ensure that this
environment variable is available there, too.
ellmer also automatically detects many of the OAuth or IAM-based
credentials used by the big cloud providers (currently
`chat_azure_openai()`, `chat_aws_bedrock()`, `chat_databricks()`,
`chat_snowflake()`). That includes credentials for these platforms
managed by [Posit
Workbench](https://docs.posit.co/ide/server-pro/user/posit-workbench/managed-credentials/managed-credentials.html)
and [Posit
Connect](https://docs.posit.co/connect/user/oauth-integrations/#adding-oauth-integrations-to-deployed-content).
If you find cases where ellmer cannot detect credentials from one of
these cloud providers, feel free to open an issue; weโre happy to add
more auth mechanisms if needed.
## Using ellmer
You can work with ellmer in several different ways, depending on whether
you are working interactively or programmatically. They all start with
creating a new chat object:
``` r
library(ellmer)
chat <- chat_openai("Be terse", model = "gpt-5.4-mini")
```
Chat objects are stateful [R6 objects](https://r6.r-lib.org): they
retain the context of the conversation, so each new query builds on the
previous ones. You call their methods with `$`.
### Interactive chat console
The most interactive and least programmatic way of using ellmer is to
chat directly in your R console or browser with `live_console(chat)` or
`live_browser()`:
``` r
live_console(chat)
#> โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
#> โ Entering chat console. Use """ for multi-line input. โ
#> โ Press Ctrl+C to quit. โ
#> โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ
#> >>> Who were the original creators of R?
#> R was originally created by Ross Ihaka and Robert Gentleman at the University of
#> Auckland, New Zealand.
#>
#> >>> When was that?
#> R was initially released in 1995. Development began a few years prior to that,
#> in the early 1990s.
```
Keep in mind that the chat object retains state, so when you enter the
chat console, any previous interactions with that chat object are still
part of the conversation, and any interactions you have in the chat
console will persist after you exit back to the R prompt. This is true
regardless of which chat function you use.
### Interactive method call
The second most interactive way to chat is to call the `chat()` method:
``` r
chat$chat("What preceding languages most influenced R?")
#> R was most directly influenced by:
#>
#> - **S** โ the main immediate ancestor; R was designed as an open-source
#> implementation inspired by S.
#> - **Scheme / Lisp** โ especially in ideas around functional programming and
#> language design.
#> - **FORTRAN** โ via numerical computing and the statistical computing
#> tradition.
#> - **C** โ for implementation and low-level language structure.
#> - **Awk / Unix tools** โ some influence on data handling and scripting style.
#>
#> If you mean the **biggest single influence**, itโs **S**.
```
If you initialize the chat object in the global environment, the `chat`
method will stream the response to the console. When the entire response
is received, itโs also (invisibly) returned as a character vector. This
is useful when you want to see the response as it arrives, but you donโt
want to enter the chat console.
If you want to ask a question about an image, you can pass one or more
additional input arguments using `content_image_file()` and/or
`content_image_url()`:
``` r
chat$chat(
content_image_url("https://www.r-project.org/Rlogo.png"),
"Can you explain this logo?"
)
#> This is the **R logo**, used for the R programming language.
#>
#> What it represents:
#> - The big **R** is the language itself.
#> - The gray oval/loop behind it is part of the stylized brand mark.
#> - The design is meant to look simple and technical, fitting a statistical
#> computing language.
#>
#> A common interpretation is that it echoes the idea of a **circle/loop** around
#> the letter, but thereโs no official deep symbolic meaning beyond branding.
#>
#> If you want, I can also explain:
#> - the **history of the R logo**
#> - what **R** stands for
#> - how it differs from the **S** language logo/history
```
### Streaming vs capturing
In most circumstances, ellmer will stream the output to the console. You
can take control of this by setting the `echo` argument either when
creating the chat object or when calling `$chat()`. Set `echo = "none"`
to return a string instead:
``` r
my_function <- function() {
chat <- chat_openai("Be terse", model = "gpt-5.4-mini", echo = "none")
chat$chat("What is 6 times 7?")
}
str(my_function())
#> 'ellmer_output' chr "6 times 7 is 42."
```
If needed, you can manually control this behaviour with the `echo`
argument. This is useful for programming with ellmer when the result is
either not intended for human consumption or when you want to process
the response before displaying it.
## Learning more
ellmer comes with a bunch of vignettes to help you learn more:
- Learn key vocabulary and see example use cases in
`vignette("ellmer")`.
- Learn how to design your prompt in `vignette("prompt-design")`.
- Learn about tool/function calling in `vignette("tool-calling")`.
- Learn how to extract structured data in `vignette("structured-data")`.
- Learn about streaming and async APIs in `vignette("streaming-async")`.
- Learn tips for programming with ellmer, including inside your own
package, in `vignette("programming")`.
## Privacy
[Privacy policy](https://tidyverse.org/google_privacy_policy/)