Your knowledge base¶

Every agent run starts from what your workspace knows. An agent with no context writes plausible marketing copy about a product it has never seen. An agent that knows your positioning, your customers and your house style writes something you can actually ship.

Context is also the only part of Questo that compounds. A task is spent once. A tenet you sharpen today improves every run for as long as it stays true.

The knowledge base has five tabs, and they answer different questions.

Tenets: what is true about your business¶

Tenets are the eight standing facts about your company:

There are always exactly eight. You cannot add a ninth or delete one, because every agent run reads all of them and a stable set keeps that reliable. What you can do is edit any of them, at any time, and an empty tenet is a legitimate state if you genuinely do not have a view yet.

Onboarding fills in the first three (product, positioning, targeting) from your answers. Read them before you do anything else. They were drafted from a short conversation, and the wording is usually close but rarely exact.

Each tenet also carries a definition: scaffolding that explains what belongs in that tenet, which you can adapt to how your team thinks. The definition is the prompt; the text is your answer.

Tenets can be shared publicly by link, which is useful for handing your positioning to an agency or a contractor.

Skills: reusable know-how¶

A skill is a piece of reference content the agent should have at hand. A messaging framework, a qualification checklist, the structure your one-pagers follow, the way your team runs a competitive teardown.

Use a skill when you find yourself pasting the same explanation into task prompts. Write it once, and it is available on every run.

Rules: standing instructions¶

A rule is an instruction the agent should follow rather than a fact it should know. "Never claim we are the market leader." "Always include a pricing line in outbound emails." "Write dates as 14 March, not 3/14."

Rules are the natural home for corrections. When you edit an output because the agent did something you always want done differently, that difference is a rule.

The distinction from skills is worth keeping clean. A skill is knowledge the agent draws on. A rule is a constraint the agent obeys.

Files: your documents¶

Files are documents in your workspace: uploads from you, and deliverables the agent produced while working. Both live here.

The switch that matters is knowledge. Marking a file as knowledge means "this is reference material for the whole workspace", and it is what the agent reads on future runs. It is curation, not history. A file being present does not make it knowledge; you saying so does.

That distinction exists because the agent writes a lot of files while it works. Intermediate research, scratch drafts, one-off exports. If every byte the agent ever wrote were treated as reference material, your context would fill with noise inside a week. So the default is no, and you promote the ones worth keeping.

Files you attach to a single task are a different thing again. Those are context for that piece of work, not for the workspace.

URLs: live references¶

A URL is a named link the agent can read as reference. Your own site and pricing page, competitor pages, an analyst report, a docs site you integrate with.

Add the pages you would check yourself before writing something. The agent will check them for you.

What is not in here¶

Insights live in their own place, not in the knowledge base, and the distinction is worth keeping.

The knowledge base is what is durably true: your positioning, your house style, the framework you use. An insight is a dated signal: what a prospect said last Tuesday, what a metric read last month. Both feed the agent, but one is standing context and the other is news.

When a signal turns out to be durable, promote it by hand. A quote from one customer is an insight. "Our buyers care more about audit logs than about price" is a tenet.

Where the agent fits¶

The agent can propose knowledge edits itself. When it learns something during a run that contradicts or extends what you have written, it can draft a change to a tenet, skill or rule.

It cannot apply that change on its own. Knowledge writes always land in your inbox as a proposal you approve or reject. This is deliberate: the knowledge base is the thing everything else reads from, so it is the last place where a silent edit would be acceptable.

The one exception is not really an exception. If you already have the knowledge tab open on the exact item the agent is editing, the change applies live and you watch it happen, because a card asking you to approve something you are staring at adds nothing.