Ask your project

The Ask tab turns a project into something you can interrogate in plain language. "What's blocking the launch?" "Which tasks are overdue?" "Summarize the open risks." sprintrr answers using only what's in the project — and cites every claim so you can verify it.

How it works

1. Open a project and switch to the Ask tab.
2. The first time, sprintrr builds a search index of the project's content (tasks, milestones, comments, status updates, risks). This happens automatically; you'll see "N indexed" once it's ready.
3. Type a question and press Ask (or Cmd/Ctrl + Enter).
4. You get a concise answer with source pills beneath it. Click a pill to jump to the task, milestone, risk, or status update it came from.

Grounded, not guessing

This is the part that matters. The assistant is instructed to answer only from your project's data — never from general knowledge, never by inventing tasks or dates. If the project doesn't contain enough to answer, it says so plainly and flags the answer as low-confidence rather than making something up. Every factual claim maps back to a numbered source.

That's the difference between a chatbot and a research tool: you can trust the answer because you can check it.

What gets indexed

• Project — name, description, dates
• Tasks — title, status, priority, due date, details
• Milestones — title, status, progress, target, description
• Comments — the discussion on your tasks
• Status updates — the 20 most recent, drafts included
• Risks — the risk register, with mitigation plans

The index is a snapshot. After you make significant edits, click Reindex to refresh it — or just reopen the tab, which re-checks automatically when the index is empty.

Cost

Ask is free — questions don't use any AI Credits, and building or refreshing the index is free too. Answers run on sprintrr's own Gemini 3.5 Flash-Lite (the Speed model in AI generation), with Claude Haiku 5.5 as the fallback if Gemini is unavailable. With BYOK (your own AI key) selected, the answer is generated on your key and model instead. Your project data is never used to train anyone's model.

Tips for good answers

• Be specific: "Which High-priority tasks are still Not Started?" beats "what's left?".
• Ask one thing at a time — compound questions get compound, vaguer answers.
• If an answer is low-confidence, the project probably doesn't capture what you're asking about yet. Add the detail as a task/comment, reindex, ask again.

Privacy

• The assistant only ever sees data from the project you're asking about.
• Retrieval is hard-scoped to that project at the database level — there's no way for one project's question to surface another project's data.
• Embeddings (the search index) always run on sprintrr's own Gemini key, even when you use BYOK — only the answering step runs on your key. sprintrr stores the embeddings (to make search fast) but never the answers.

Roadmap

• Streaming answers with inline citation markers.
• Automatic reindex on entity changes (today it's on-demand).
• Cross-project / portfolio questions.
• Uploaded documents (PDFs, specs) as additional grounding.