askLLM 1.3.1 — full-tutorials link in results, card, and README

askLLM 1.3.1 is a small follow-up to 1.3.0 that surfaces the companion statistics tutorials throughout the module. No behaviour changes to the two analyses.

What's new

The full statistics tutorials — the sister project stat-skills-tutorials — are now linked from every natural entry point:

• Both analyses' results panel — a clickable "Statistics tutorials (full course)" link at the top, shown the moment you open the analysis.
• On-open instructions — a bilingual (EN/中文) pointer line for learners.
• The module card in jamovi's Modules manager — the tutorials URL now appears in the module description.
• The README (EN + 中文).

askLLM's own learn-r and choose-model pages are kept as quick reference, with the full course as the primary destination.

Everything else — two analyses (jamovi Module Guider + R code tutor), the summary-statistics-only privacy design, and menu/package grounding — is unchanged from 1.3.0.

Install

Download askLLM_1.3.1_win64.jmo below → in jamovi, click ⊕ (top-right) ▸ Side-load ▸ choose the file. Windows 64-bit, tested on jamovi 28.2.0.0. Bring your own API key (NVIDIA / Gemini / OpenRouter / GitHub Models / Ollama / custom); see https://scgeeker.github.io/askLLM/choose-model.html

askLLM 1.3.0 — two analyses: jamovi Module Guider + R code tutor

askLLM 1.3.0 splits the module into two analyses, both under Analyses ▸ askLLM.

jamovi Module Guider

"Which jamovi analysis should I run?" — recommends the right analysis and quotes the exact menu path, grounded in the modules you actually have installed. If you ask for R code, it points you to R code tutor.

R code tutor (new)

"How do I write the R for this?" — writes R that works on your dataset for you to paste into the Rj Editor and run yourself (the tool teaches, you execute). Grounded in the R packages actually bundled with your Rj environment, so it only uses what you have — it never suggests install.packages(), read.csv(), or file paths. Three personas: Consultant (a ready snippet), Tutor (a # TODO skeleton), Explainer (line-by-line comments).

Rj Editor is a module for the desktop version of jamovi (not jamovi Cloud), so R code tutor needs desktop jamovi.

askLLM is your copilot, not an autopilot

Both analyses only advise: analysis strategies, jamovi menu paths, and R code you can paste into the Rj Editor. You run the analysis, you write to any column, you drive the jamovi interface — askLLM does none of that for you. This is now a stated principle, not just a description of the current build.

For a multi-step analysis, iterate like this: summarise the previous step's results in your next question, and the LLM builds its suggestion for the next step on that. askLLM does not read jamovi's analysis output (a platform limitation, not a missing feature), so you are the one who carries each result back.

Also new

• Learn R with Rj — a bilingual (EN/中文) guide page: https://scgeeker.github.io/askLLM/learn-r.html
• Each analysis's results link to the learn-R guide and a model picker.
• Bidirectional prompt boundaries: each analysis stays in its lane and redirects off-topic questions to the other.
• The experimental "actions" mode was removed from the UI (agent-style acting belongs in an upstream automation API, not an analysis module); its code is retained, dormant. Under the copilot boundary above, that path either stays dormant or, if it is ever re-wired, must still leave execution to the user.

Privacy (unchanged, by design)

Only summary statistics of the variables you pick are sent — never raw data rows — plus your question and (for R code tutor) the names of your Rj packages. askLLM never runs R for you. With a local model (Ollama), nothing leaves your machine.

Install

Download askLLM_1.3.0_win64_jamovi-2.7.jmo below → in jamovi, ⋮ (top-right) ▸ modules ▸ install from file / sideload. Bring your own API key (NVIDIA / Gemini / OpenRouter / GitHub Models / Ollama / custom); see the in-app help and https://scgeeker.github.io/askLLM/choose-model.html

---

Notes updated 2026-09-10: added the copilot boundary statement, and corrected a documentation error in LIMITATIONS where an illustrative example of hallucination had been described as a recorded test result. The module itself is unchanged from the 1.3.0 build below.

askLLM 1.1.1 — honesty and grounding fixes

Patch release — honesty and grounding fixes found by re-examining the v1.1 test screens.

修正版——重新檢查 v1.1 測試畫面後發現的誠實性與 grounding 修正。

Fixes / 修正

• The verification caveat no longer lies when the module list is off. With Include installed modules unticked (or scanning failed), the caveat now says menu paths were NOT verified, instead of claiming they were checked. / 關閉「Include installed modules」時,查證提醒改為明說「選單路徑未經比對」,不再謊稱已比對。
• Tighter module-name grounding. The instruction to the model changed from "never invent module names" (which failed — the model doesn't consider citing a real off-list module to be inventing) to "name ONLY modules that appear literally in the provided list". A live re-test confirmed the previously observed off-list leak (a real module not in our list) is gone. / 反洩漏措辭從「不要編造模組名」改為「只能提清單裡逐字出現的名字」;live 重測確認先前的清單外洩漏已消失。

Residual, documented honestly: the model may still mention R packages (e.g. lavaan) when suggesting an Rj-based workaround — that is a legitimate reference, not a fabricated module. / 誠實記錄的殘餘:模型仍可能在建議 Rj 變通法時提到 R 套件(如 lavaan),那是合理引用而非虛構模組。

Install / 安裝

Download askLLM_1.1.1_win64_jamovi-2.7.jmo → jamovi ⊕ → jamovi library → Side-load. Windows x64 + jamovi 2.7 only.

SHA-256: ACB63A4016BB31FF2EB025FC715839593026ED4D9F5D2147ADF5325A9581812E

529 unit tests passing.

askLLM 1.1.0 — module-aware statistical consultant

askLLM 1.1 makes the LLM cite only real menu paths. The module now scans the jamovi modules actually installed on your machine, attaches the real menu tree to the prompt, and instructs the model to recommend only from it — suggesting installs only from the official jamovi library list otherwise.

v1.1 讓 LLM 只引用真實選單路徑。 模組會掃描你機器上實際安裝的 jamovi 模組,把真實選單樹附進 prompt,並指示模型只能從中推薦;清單外的需求只能從官方 jamovi library 清單建議安裝。

Why / 為什麼

In v1.0 testing every model invented jamovi menu paths — including menus that do not exist. With the catalog attached, an A/B test over the same questions scored 18/18 cited paths quoted verbatim from the real menu tree (100%, zero fabrications). Details: docs/LIMITATIONS.*.md and dev-notes/catalog-hit-rate.md.

v1.0 實測中所有模型都會編造選單路徑(連不存在的選單都寫得出來)。附上目錄後的 A/B 對照:引用的 18 條路徑 18 條逐字命中真實選單(100%,零虛構)。詳見 docs/LIMITATIONS.*.md 與 dev-notes/catalog-hit-rate.md。

What's new / 新內容

• Scans installed modules (builtin + user side) and grounds suggestions in the real menu tree / 掃描已安裝模組(內建+使用者側),建議以真實選單樹為據
• Suggests uninstalled modules only from the official jamovi library listing (synced at release time) / 未安裝模組只從官方 library 清單建議(發佈時同步)
• New option Include installed modules (on by default; untick to revert to v1.0 behaviour) / 新選項 Include installed modules(預設開;關閉即回 v1.0 行為)
• Privacy disclosure updated: the installed-modules list is environment metadata only — no data values — and can be switched off / 隱私聲明更新:模組清單屬環境中繼資料,不含資料值,可關閉

Install / 安裝

Download askLLM_1.1.0_win64_jamovi-2.7.jmo below → jamovi ⊕ → jamovi library → Side-load → choose the file. Windows x64 + jamovi 2.7 only.

下載下方 .jmo → jamovi 右上 ⊕ → jamovi library → Side-load 分頁 → 選擇檔案。僅限 Windows x64 + jamovi 2.7。

SHA-256: F3AFAE182E525C1BBD8976E36B4B1CB2E05326BB99680D10953274A20346C9AF

520 unit tests passing; SDD spec with 16 acceptance scenarios in specs/.

askLLM 1.0.0

Ask an LLM about the dataset you have open in jamovi. Select variables, type a question, and the module attaches their summary statistics — never the raw rows — so the answer is grounded in your data, often with concrete jamovi menu paths for the analysis it suggests.

在 jamovi 裡直接問 LLM 關於「你的資料」的問題。勾選變項、輸入問題,模組會附上這些變項的摘要統計(不含原始資料列),讓 LLM 針對你的資料集回答,並常常給出具體的 jamovi 選單路徑建議。

Install / 安裝

Download askLLM_1.0.0_win64_jamovi-2.7.jmo below, then in jamovi: ⊕ (top right) → jamovi library → Side-load tab → choose the file.

下載下方的 .jmo,在 jamovi 中點右上角 ⊕ → jamovi library → Side-load 分頁 → 選擇該檔案。

This build targets Windows x64 + jamovi 2.7 (current series). A .jmo only works on the OS, CPU architecture, and jamovi series it was built on.

本檔為 Windows x64 + jamovi 2.7 專用。.jmo 只能安裝在建置時對應的作業系統、CPU 架構與 jamovi 系列版本。

Providers

Provider	Free tier	Runs	Setup
NVIDIA NIM	yes, no card	cloud	SETUP-nim
Google Gemini	yes, no card	cloud	SETUP-gemini
GitHub Models	yes, GitHub account	cloud	SETUP-github
Ollama	free, no key	your machine	SETUP-ollama
Custom	—	any OpenAI-compatible endpoint	SETUP-custom

Privacy / 隱私

Only summary statistics of the variables you select are sent — never raw data rows. API keys are read from your environment variables (or .Renviron) and are never written into the .omv file. Use Ollama for zero data leaving your machine.

只有你所選變項的摘要統計會被送出,原始資料列不會。API 金鑰從環境變數(或 .Renviron)讀取,不會寫入 .omv 檔。若不希望任何資料外送,可改用 Ollama(本機執行)。

Notes

• Verified in jamovi 2.7.37 against NVIDIA NIM and Google Gemini.
• Built with jmvtools; bundles ellmer 0.2.0 from the jamovi CRAN snapshot.
• 198 unit tests passing.