Adrails Docs

Knowledge and skills

What the agent works from in your workspace, from the memory it keeps of your chats to your documents, skills, shortcuts and your store's brand.

Knowledge holds what the agent works from in your workspace. Its tabs are Overview, Memory, Documents, Skills and Brand. Overview shows the first lines of each of the other four, with their count and the way to all of them. Add, at the top, writes a document, uploads a file, creates or imports a skill, creates a shortcut or adds a Shopify store; on a tab, it adds only what that tab holds.

Memory: what the agent keeps

Memory is one table of what the agent keeps and uses in your chats, with a search, a sort by most used or newest, and three filters:

  • Instructions: how the agent works in every future conversation, such as report every figure in the ad account's currency. State one in a chat and the agent keeps it, or choose Add instruction.
  • Results: what was measured, with its figures, where and over which dates. The agent starts the next brief from it. Add one from the menu beside Add instruction.
  • Profile: your business facts, from onboarding and your chats, read in every answer.

Each line says where it came from, such as a chat or a split test, and how often the agent used it. Open a line to edit it, or Forget it. What the agent learned of an ad account, such as its target or how its ads are named, is kept here too.

Documents

Write a document, or upload a PDF, Markdown, text or CSV file. The agent reads a document when a request calls for it. New folder groups documents, and a folder opens as its own page. Filter by Written, Uploaded, Generated, Drafts or Archived, search, and show them as a list or a grid. Generated holds what Adrails wrote, such as reports.

Ask for a report in a chat, such as a report of the last 7 days, and the agent delivers it as a document in Documents: its analysis, then the tables of its reads, with the date range above them. The chat keeps a short answer and the link. A quick figure, such as yesterday's ROAS, or a report asked here in the chat stays in the chat. Ask for a doc of any answer, such as make it a doc, and the agent saves one and links it too. It holds the agent's text, then every table and chart of its last answer as tables, with the figures the reads returned. Ask for a table per day and the answer lists each day's ad spend, ROAS and attributed revenue, with the period's total as its last row. A document made in a chat stays as private as the chat.

Skills

The Skills tab has three views, each with a search: Skills, Shortcuts and Archived.

Skills: reusable methods

A skill tells the agent how your team handles a task, such as a creative review or a budget pacing check. The agent looks up the relevant skills on its own when a request calls for one.

The Skills view lists your workspace's own skills, then, under Already included, the five that come with Adrails: Meta ad copywriter, Campaign builder, Creative strategist, Performance analyst and Budget planner. An included skill is on with no action needed: open it and choose Duplicate to make your own version, or Try now to use it in a chat. Custom agents your workspace made are listed with its skills.

To add a skill:

  1. Open Knowledge, then Skills, and choose Create skill. Import makes one from a Markdown or text file instead.
  2. Name it, and give it a Key: 2 to 32 lowercase letters, numbers or hyphens.
  3. Choose its Meta accounts and Integrations: all of them or selected ones. A skill is used only when its scope covers the whole selection of the request.
  4. Say When to use it, in up to 180 characters, and write its instructions in SKILL.MD, from 30 to 50,000 characters. Markdown is allowed.
  5. Choose Create skill.

Saving an edit raises the skill's version, and Undo on the confirmation puts the previous text back. Archive moves a skill or a shortcut to Archived, where the agent no longer reaches it, and Restore brings it back.

A skill guides the agent; it grants nothing. The agent reads it as a workspace procedure, below your request and Adrails' own rules. It cannot approve or apply a change.

Shortcuts: saved requests

A shortcut is a request you send often. Type / in the message box, pick the shortcut, review it and send it. Pin a shortcut to keep it at the top.

Create one in the Shortcuts view, or from Add: a name, a command, and the request, from 30 to 4,000 characters. In a shortcut, {{input}} is replaced by what you type after the command. Without it, what you type is added as an extra instruction.

A new workspace starts with 4 shortcuts: /us, /daily, /creative and /store. Typed with your store's address, /store reads the site and opens the launch card with what it found.

Sending a shortcut is one AI message, counted in credits like any other: see The agent and credits.

Specialist agents

Pick a specialist for a conversation from the agent menu in the message box, or leave it on Automatic: Campaign Manager, Creative Strategist, Analysis Expert, Media Buyer, Copywriter, Store & Offer or Reporter.

Brand

Brand holds the brand of each store, in nine sections: Overview, Story and mission, Products, Audience, Positioning, Voice and tone, Promises, Objections and answers, and Claims to avoid. Read my store drafts them from the connected Shopify store, or from an address you give. Review each draft to confirm it, edit it or clear it. When the store changes, a confirmed section offers its newer draft, and you keep yours or use it. Every ad copy the agent writes follows the confirmed sections.

Who can change what

What you can do in Knowledge follows your role's Knowledge and skills permission, where reading, creating, editing and deleting are each granted on their own. Owner, Administrator and Member include all of them; a custom role has what you gave it. A button your role cannot use is disabled and says why: Your role can't create knowledge and skills. Ask an admin.

Watch for

  • A skill you delete stays deleted: it is not added back.
  • A skill scoped to some accounts is not used when the request also covers other accounts.
  • An archived skill or shortcut is kept, but the agent no longer reaches it.

On this page