Recall is a local retrieval workspace: it builds a searchable index of the things you have already written down, and answers questions against it with citations. In the app itself the action is still named Ask Memory, which is the older name and the more accurate one.
What it is
Trigger it with recall, ask memory, ask notes, or knowledge. Ask a question in ordinary language. Vehla searches the local index, pulls the passages that look relevant, and hands them plus your question to whichever model you have selected. The answer comes back with numbered citations pointing at the source passages.
The indexing and the retrieval are always local. That part does not depend on your model choice or your network.
What it indexes
Only the sources you explicitly enable in Settings. Nothing is indexed by default that you did not turn on.
Note what is not on that list: your email, your Slack, your browser history, your calendar, your messages. Recall is not a system-wide surveillance index of your Mac. It searches the things you deliberately put somewhere.
What it's good at
The queries that work are the ones where you know the information exists and cannot remember where you put it. "What did we decide about the launch date." "What was the connection string for staging." "Which of these three vendors did I say was too expensive."
The clipboard-history source is the sleeper here. Half the useful answers in a normal week come from something you copied and never saved anywhere — a URL, a key, a paragraph from a document you closed. It is the only part of your working memory that gets captured without any discipline on your part.
It is also good at the question you would otherwise not bother asking, because searching four apps manually costs more than the answer is worth. That is the actual gain: not better answers, but asking more often.
What it's bad at
Questions whose answer you never wrote down. Retrieval cannot invent what is not in the index. If the decision happened in a meeting and nobody took notes, Recall will either say so or — worse — assemble something plausible from adjacent passages. Read the citations.
Counting and aggregating. "How many times did I mention this client" is the wrong shape of question. Retrieval finds passages, not totals.
Anything time-sensitive across many documents. "What changed between the March and June versions" requires reasoning over the whole corpus, not a handful of retrieved chunks. A small local model will do this particularly badly.
Exact-string lookups. If you know the precise phrase, ordinary search in the Notes workspace is faster and more reliable than asking a model about it.
Citations are the feature
Every answer carries numbered citation chips. Each chip can open the source, reveal it in Finder, copy its content, or stage a supported file on Shelf. This is not a trust badge — it is the working part of the interface.
The right way to use Recall is to treat the prose answer as a lookup mechanism for the citations and then read the citation. A retrieval answer with no source is a guess with good grammar. With sources, the answer being slightly wrong costs you nothing, because you are one click from the original sentence.
The retrieval split
This matters and it is easy to get wrong. Indexing and retrieval are local, unconditionally. The answer is generated by whichever model you selected.
Pick a local model and the whole chain stays on your Mac. Pick a cloud model and the assembled prompt — your question plus the retrieved passages from your notes and clipboard — goes to that provider. "Retrieval is local" is a true statement that does not mean "your notes stay on your Mac" unless you also chose a local model. Our guide to what actually stays on the device maps the same boundary across every Vehla feature.
Tuning your sources
More sources is not better. Every enabled source adds passages that can be retrieved instead of the right one, and the failure mode of a noisy index is confident answers assembled from the wrong material.
What worked for us: notes and clipboard on, always. Snippets on — they are short and high-signal. Canvases off, because a drawing's extracted text is rarely the answer to a question. Kanban on only if you actually run your work through the board. Selected folders scoped tightly to a project directory rather than pointed at Documents.
If answers start feeling vague, the first thing to try is turning a source off rather than switching to a bigger model.
The habit it needs
Recall's quality is a function of what you put in front of it, which means it is the one feature in Vehla that requires a small behavioral change: capturing things. Quick Capture into the daily note, bound to a hotkey, is the whole discipline. One line, no formatting, no filing.
People who install Recall expecting it to know things immediately are disappointed for about two weeks and then find it indispensable, which is an unglamorous thing to admit on a marketing site. If you want the version of this that needs no new habits, the clipboard source is already collecting.
Related: what actually stays on the device, and the commands that actually get used.
Common questions
What does Recall index?
Only the sources you enable: notes, clipboard history, Shelf items, snippets, canvases, Kanban boards, Deep Research reports, and selected folders. Not your email, Slack, browser history, calendar, or messages.
What is Recall bad at?
Questions whose answer you never wrote down, counting and aggregating, comparisons across many documents, and exact-string lookups — where ordinary search in the Notes workspace is faster and more reliable.
Should I enable every Recall source?
No. Every source adds passages that can be retrieved instead of the right one, and a noisy index produces confident answers built from the wrong material. If answers feel vague, turn a source off before switching to a bigger model.