NEWS

---

rollama 0.3.1.9000

• query() cache file names are now more stable and survive different locales and R versions. Settings that do not change the response, like keep_alive and stream, are no longer used to produce the hash. Failed requests are retried only up to 3 times now (see options(rollama_cache_retries)) instead of indefinitely, then an error lists the reasons. Identical requests are sent only once, interrupting no longer restarts unfinished requests, and the cache directory is created for single requests too.

rollama 0.3.1 (2026-08-24)

• query() can now batch annotate several images with the same prompt by passing images as a list (one element per query)
• updated image annotation vignette to explain batch annotation
• create_model() gains renderer and parser params
• query()/chat() output = "data.frame"/"list" now include thinking and tool_calls from the response
• think argument in query()/chat() now accepts "high"/"medium"/"low"/"max" in addition to TRUE/FALSE
• fixes issue in tests
• bug fixes

rollama 0.3.0 (2026-03-25)

• adds option to cache responses
• query now supports logprobs output
• make it possible to supply several questions at once
• add authentication vignette
• implemented structured outputs (including new vignette)
• synced package with Ollama API changes
• adds list_running_models()
• updated chat() and query()
• update parameters in embed_text()
• update parameters and output in show_model()
• rewrote progress and answer streaming
• bug fixes

rollama 0.2.1 (2025-04-24)

• added support for structured output
• added support for custom headers (e.g., for authentication)
• added option for custom outputs
• some bug fixes and improved documentation

rollama 0.2.0 (2024-12-06)

• added make_query() function to facilitate easier annotation
• added more output formats to query()/chat()
• improved performance of embed_text()
• improved performance of query() for multiple queries
• changed default model to llama3.1
• added option to employ multiple servers
• pull_model() gained verbose option
• improved annotation vignette
• added vignette on how to use Hugging Face Hub models
• some bug fixes

rollama 0.1.0 (2024-05-01)

• adds function check_model_installed
• changes default model to llama3

rollama 0.0.3 (2024-03-21)

• add option to query several models at once
• dedicated embedding models are available now (see vignette("text-embedding", "rollama"))
• error handling and bug fixes

rollama 0.0.2 (2024-01-29)

• Initial CRAN submission.