How to ask FarkiBot about your model

Ask is FarkiBot’s default task. It answers questions about the models in your project — structure, gaps, relationships, summaries, inconsistencies — from the model itself.

Before you start, set up FarkiBot with an API key.

Ask a question

Open View > FarkiBot (Ctrl+Shift+K)
Leave the task dropdown on Ask
Type your question and press Enter

Shift+Enter inserts a new line instead of sending. Stop cancels mid-answer.

Follow-up questions keep the conversation, so you can drill in without restating the context. Reasoning models show their thinking while they work.

What FarkiBot can see

The Context chip in the toolbar shows exactly what will be sent: your project’s packages, exported as Markdown, minus the MOF core. Click it to narrow the scope to specific packages.

Narrowing is worth doing when:

The project is large and the question is about one part of it
You want a cheaper answer — context is most of what you pay for
You want the answer to ignore a package that would otherwise muddy it

Questions that work well

Kind of question	Example
Structure	“What application components support the order-to-cash process?”
Gaps	“Which business capabilities have no supporting application?”
Consistency	“Are there elements with no relationships at all?”
Summary	“Summarise the technology layer for someone who has never seen this model.”
Impact	“If FQ Vessels were retired, what would be affected?”
Coverage	“Which elements are missing a Lifecycle Status value?”

The model is structured data, so questions about counts, coverage and connections tend to be answered precisely rather than approximately.

FarkiBot answers only from your model

By default FarkiBot answers only from your model and your attachments. If something isn’t there, it says so rather than filling the gap from memory.

That’s deliberate, and it’s the single most important thing to understand about the feature. An answer you can trace back to named model content is worth something; a plausible answer assembled from an LLM’s training data is worth nothing when you’re about to make a decision with it. See How FarkiBot Works.

If FarkiBot tells you the model doesn’t say, that’s usually a finding about the model, not a failure of the answer.

Attach supporting files

Click the paperclip to add .md or .txt files — a glossary, meeting notes, a standard you’re checking against. Attachments join the model as context for the question.

For generation tasks, more file types are accepted. See How to generate a model.

Turn on web research

Some questions need the outside world: how does our portfolio compare to current practice, are these technologies still supported, what does this standard require now. Click Web research off below the question box to turn it on; the button then reads Web research on.


Off by default, chosen per task	A question about your own model should stay grounded. Research is a deliberate act, and turning it on for Ask does not turn it on for Report or the generation tasks.
You are told while it is on	A notice sits above the question box for as long as research is on, and the send button reads Ask with web research.
Your model stays authoritative	Web results are evidence about the world, compared against your model — never a replacement for reading it, and never reported as being in your model.
Every web claim is cited	As a link you can follow. If the search couldn’t answer something, FarkiBot says so instead of guessing.
It costs more	Often several times a plain answer: the LLM runs several searches and reads the results. Answers that used research carry a Web badge.
Not every model can	In the LLM list, models that can research carry the researcher icon. On OpenRouter that is every model that supports tool use, which is most of them. Anthropic and Google support it directly on current models. A local or custom endpoint can’t. With research on, models that can’t are listed last, and choosing one turns research off and tells you.

What leaves your browser: your provider runs the search, so the search terms the LLM writes — which may name things from your model — reach a search engine. Your model export itself is never sent to a search engine, and Farkitect’s servers are not in the path either way.

When an answer finishes

A summary opens when the run finishes: what FarkiBot did, the LLM it used, how long it took, the tokens it used, what it cost, and whether it used web research. Click OK to close it.

The essentials stay under each answer on one line, for example moonshotai/kimi-k3 · 14,668 tokens · $0.02. Click that line to open the summary again.

OpenRouter reports cost; Anthropic and Google don’t. Those runs show “cost not reported”, and you should check your provider console for spend.

Copy, save and clear

Three buttons sit under each answer, and the same three sit at the right of the toolbar for the whole conversation.

Button	Under an answer	In the toolbar
Copy	Copies that answer, with its Markdown formatting	Copies the whole conversation
Copy as plain text	Copies that answer with the formatting removed	Copies the whole conversation as plain text
Save	Saves that answer as a Markdown file	Saves the whole conversation as one Markdown file

The whole conversation is laid out as your questions and FarkiBot’s answers in order, each answer naming the LLM that wrote it. Answers that failed or were stopped are left out.

Clear chat, the eraser button beside them, empties the conversation and starts again. It does not ask first.

Your model choice is remembered per task, so Ask and the generation tasks can each keep their own.

Turning an answer into a document

If what you actually want is something to hand over rather than something to read, switch the task to Report. See How to write reports with FarkiBot.