NEWS

---

PacketLLM 0.1.3 (2026-06-29)

• Attachments: add support for .csv files. CSV data is converted into a model-friendly representation consisting of a dimensions header, a column schema with inferred types, and the rows rendered as a Markdown table.
• Attachments: the CSV field delimiter is detected automatically (comma, semicolon, tab, or pipe), and semicolon files with comma decimal marks are handled.
• Attachments: large CSV files are truncated to the first 1000 rows with an explicit note; over-long cell values are shortened.
• Docs: update README, vignette, and the in-app help to list .csv as a supported attachment format.

PacketLLM 0.1.2 (2026-06-25)

• UI: add onboarding help modal (shown once on first launch, reopenable via Help button) covering context modes, editor actions, settings, and persistence.
• UI: show model ID alongside preset name in the selector (e.g. "Best - gpt-5.5").
• UI: add visible "Context:" label before the context mode selector.
• UI: render the assistant label in accent color for easier chat scanning.
• UI: grey out the custom instruction field when behavior is not set to Custom.
• Context: replace manual refresh button with automatic 1-second editor polling; context updates silently when selection or file changes.
• Settings: fix custom instruction being silently ignored when behavior preset was not set to Custom.
• Code: consolidate internal %||% helper to a single definition.
• Code: replace internal getFromNamespace reflective lookup with a direct reference.
• Docs: fix vignette install.packages() call (missing quotes); add usethis and remotes to Suggests.

PacketLLM 0.1.1 (2025-08-23)

• Models: replace old built-in defaults with model presets (Best, Balanced, Fast) backed by gpt-5.5, gpt-5.4, and gpt-5.4-mini.
• RStudio workflow: add editor-aware context capture, Addin registration, and explicit copy, insert, and replace actions.
• UI/Settings: redesign the gadget with responsive code-focused output, assistant behavior presets, reasoning effort, verbosity, and optional max output controls.
• Safety: render model and user text safely instead of treating arbitrary chat content as raw HTML.
• Rendering: add safe Markdown and LaTeX math formatting for assistant output.
• History: save gadget conversations locally so they can be restored after closing and reopening the gadget.

PacketLLM 0.1.0 (2025-04-24)

• Initial release to CRAN.
• Provides an RStudio Gadget (run_llm_chat_app()) for interactive AI-assisted development.
• Supports multiple, tabbed conversations.
• Allows uploading .R, .pdf, and .docx files as context.
• Enables setting the model and assistant instructions per conversation before the first message.
• Features asynchronous API calls to keep the R console responsive.
• Includes helper functions for managing conversation history and settings (create_new_conversation, get_active_chat_history, set_conversation_model, etc.).
• Requires an OPENAI_API_KEY environment variable for API access.
• Added basic unit tests using testthat.
• Added README.md and an introductory vignette (vignettes/PacketLLM-introduction.Rmd).