Intelligence
Which models exist, which keys back them, what the assistant is told, and which skills and prompts are offered.
Five screens decide what the assistant is, what it may do, and what it costs per turn. This is the group where a small change has a large effect on both quality and spend.
Models
Intelligence → Models is the catalogue. It is not only a list of names: it decides how the picker inside the
product is shaped.
| Setting | What it changes |
|---|---|
| Tiers | The three groups the picker leads with, and what each is called |
| What is in each tier | Which models the quick picks actually call |
| Live | A model switched off disappears from the picker everywhere |
| Context and output ceilings | How much a turn may carry, and how long one answer may be |
| Temperature ceiling | The upper bound anybody, including the agent, may ask for |
| Capability tags | Which models are trusted with tools, images or long context |
| Default tier | What a brand-new account starts on |
| Blocked list | Gateway models a user may never pick, even when the gateway offers them |
The picker shows the tiers first and the full gateway list second, filtered by this screen and by the plan's
entitlements. A model visible here and not included in a plan is refused at the plan level, and
Money → Entitlements is where that is decided.
Providers & keys
Intelligence → Providers & keys holds the keys and reports their health.
Each provider is listed with its environment name and whether a key is set.
Paste a key beside the provider and save. It is sealed with the
installation's SETTINGS_SECRET and never shown again.
Press the health check on the row to see whether the key actually works, rather than whether it exists.
Choose which providers a user may pick from, and which are reserved for the installation's own use.
A key that is already in the environment is reported as set, and the panel will not overwrite it from here. That is the right shape for a host whose secrets are managed outside the panel. See Model keys for the names and for what each provider is for.
Agent behaviour
Intelligence → Agent behaviour is what the assistant is told, and what it is allowed to do.
| Setting | What it decides |
|---|---|
| Global prompt | The standing instruction every turn starts from. The most powerful field in the panel |
| Tools per plan | Which plans may use file writing, search, images or the terminal |
| Guardrails | What the assistant must refuse, and what it must ask before doing |
| Reasoning | How much thinking a turn is allowed before it answers, per tier |
| Compaction | When a long conversation is summarised to stay inside the context window |
The global prompt is a contract with your users
Everything in it is applied to every turn for everybody. A sentence added here
changes the product's personality, its refusals and its honesty in one edit,
and it is the field most likely to be changed at two in the morning and
forgotten. Change it deliberately, and use System → Audit log to find out
who did.
Skills
Intelligence → Skills is the library of skills, and which plans may use them.
| Part | What it is |
|---|---|
| Bundled skills | The ones that ship with the installation |
| Added skills | The ones you add, with their own instructions and files |
| Per-plan access | Which plans can see a skill at all |
A person can also have skills of their own, which they manage in Customize → Skills inside the product. The installation's skills are the floor; a person's own are theirs.
Prompt library
Intelligence → Prompt library is the set of suggested prompts a new account sees, grouped into categories and
assigned per plan. It is the cheapest way to make an empty product feel like a product with a point of view:
four good starting prompts do more for a first impression than any amount of settings.
The order to change these in
Providers & keys first. Without a working key nothing else can be judged.
Models second. Decide what people can reach before deciding what they should be told.
Agent behaviour third, in small changes, testing a turn between each one.
Skills and Prompt library last, because both are content rather than configuration.
Entitlements decide what a plan unlocks, not this screen
Whether a plan may use a heavy model, web search, generated images or a large
context is decided in Money → Entitlements. This group decides what exists;
that one decides who may reach it.