Sherpa’s chat interface atDocumentation Index
Fetch the complete documentation index at: https://mintlify.com/tudoumono/Sherpa/llms.txt
Use this file to discover all available pages before exploring further.
/ui/chat.html is the primary way to explore your internal knowledge base. Type a question in plain language and Sherpa selects the right approach automatically, searches the registered document tree, and returns a structured answer with cited evidence you can download. The screen is split into three panes: a conversation list on the left, the main answer area in the centre, and a live “thinking” trace on the right that shows how the AI searched and reasoned through your documents.
Knowledge retrieval toggle
Near the message input box you will find the ナレッジ参照 (Knowledge retrieval) toggle. It defaults to off.| State | Behaviour |
|---|---|
| Off (default) | Sherpa replies as a plain conversational assistant — no document search, no citations, no scope filtering. Good for general questions, brainstorming, or quick clarifications. |
| On | Sherpa searches your registered document folders before answering. The three lenses (impact analysis, troubleshooting, spec Q&A), citations, and scope selection all become active. |
The three lenses
When knowledge retrieval is on, Sherpa classifies each question into one of three lenses and picks the appropriate analysis strategy. You do not choose the lens yourself — Sherpa infers it from the wording of your question.Impact Analysis
When to use: You are about to make a change and want to know what else it touches.Example question: 消費税率を変えると何に影響する？ (What is affected if we change the consumption tax rate?)The answer leads with a summary card: the total number of affected items broken down by certainty (sure / needs review / confirmed unaffected). Each item shows the dependency chain — for example, tax rate → tax calculation rule → estimate function — so you can trace exactly why something is flagged. You can drill into any item with a follow-up question, and you can dismiss false positives inline.
Troubleshooting
When to use: Something has gone wrong and you need to find the cause.Example question: 夜間バッチが異常終了した。原因は？ (The nightly batch terminated abnormally. What is the cause?)Sherpa returns a ranked list of cause candidates. Each candidate includes an explanation of the related components and call graph, a link to similar past incidents where available, step-by-step investigation guidance drawn from operational documentation, and source citations.
Spec Q&A
When to use: You need to understand how something works or how a value is calculated.Example question: 請求額はどう算出している？ (How is the billing amount calculated?)Sherpa returns the relevant excerpt plus a reference to the source document. If the spec exists in multiple places, all relevant passages are collected and cited.
Citations and source downloads
When knowledge retrieval is on and evidence is found, every answer ends with a citations footer. Click the 📄 icon next to any citation to download the original file — the actual Excel workbook, Word document, PowerPoint presentation, COBOL source file, or Markdown document that was registered in Sherpa. You are not downloading a converted copy; you are downloading the original binary or source.When no supporting evidence is found, Sherpa does not fabricate an answer. It returns an explicit 確証なし (“no evidence”) notice and omits the citations footer entirely. You can trust that the absence of a citation means the answer is not grounded in your documents.
Streaming responses and the thinking trace
After you send a message, Sherpa streams its progress in real time:- A thinking animation appears in the right pane showing the AI’s search process — grep passes, file reads, graph traversals, and query refinements as they happen.
- Once the analysis is complete, an answer card appears in the centre pane with the conclusion at the top, followed by breakdowns, dependency chains, or excerpts depending on the lens.
- The citations footer appears at the bottom of the answer card with download links to each source document.
AI model selector
The 頭脳 (Brain) badge near the message input lets you switch the AI provider that handles your query. The selected provider is saved to your user profile and persists across sessions. See Scope & AI for a full description of each provider and how to configure API keys.Conversation history
All conversations are automatically saved and associated with your user account. The left pane lists your own conversations and any conversations that have been shared with you. Shared conversations are read-only. You can:- Rename a conversation by clicking its title in the list.
- Pin a conversation to keep it at the top of the list.
- Resume any past conversation by selecting it — the full message history and the answer cards are restored exactly as you left them.
- Share a conversation with teammates using the share button. See Sharing for details.