Skip to main content
Knowledge holds your team’s definitions so that Addison applies them rather than inferring its own. Define “net revenue” once, and every answer that follows uses that definition. Knowledge has two parts: guides for written rules and background, and metrics for governed numbers. Both can be shared across your organization or scoped to a project, and Addison reads them before it builds anything: a chat answer, a report, a dashboard, or a deck. The fastest way to fill it is to ask. Add → Create with Addison takes a sentence or a file and hands back a draft to review and save. Write one by hand only when you already know the exact wording.
The Knowledge page at /knowledge, titled 'Knowledge' with the subtitle 'Global context available across all projects. To add Knowledge, create or modify in a project and publish.' Guides and Metrics tabs sit above an empty state: 'No Guides yet, Create or edit Guides in a project, then publish to be shared here,' with a + Add button.

The Knowledge library, shared across your organization

Guides

A guide is a short document that tells Addison how your business works: a rule, a convention, or background that changes how a number should be read. If you find yourself re-explaining something in chat, it belongs in a guide. Write it as an instruction, not a description. “Use fiscal weeks starting Monday,” not “Our company has fiscal weeks.”

What to put in a guide

The guides worth writing first answer a question your team fields every month:
  • Your fiscal calendar. When the fiscal year starts, where period boundaries fall, whether weeks run Monday to Sunday.
  • When a number is final. “Revenue before the fifth business day is preliminary and will move.”
  • What’s in and what’s out. Test accounts, internal orders, intercompany transactions, refunded lines, subsidiaries excluded from the consolidated view.
  • Segment and hierarchy definitions. What makes an account Enterprise rather than SMB, which countries roll up into EMEA, which sites belong to which region.
  • Currency handling. Your reporting currency, the rate source, and where in the pipeline conversion happens.
  • Events that bend the numbers. A pricing change in March, a warehouse migration that broke two weeks of fulfillment data, an acquisition that reset the customer count.
  • A query your team has already validated. Paste the SQL with a sentence on what it answers.
Keep each guide to one topic. Two topics means two guides, so each stays short enough to read in full and specific enough to surface on the right question.

Create a guide

Three approaches, depending on where the context already lives.
  1. Ask Addison. Describe the rule in a sentence: “Add a guide about our Q4 strategic initiatives.” Addison checks whether a guide already covers it, then drafts the body for you to review and save. You can attach a file to the same prompt (a glossary, a close checklist, an analyst onboarding doc) and ask for guides from what’s in it.
  2. Connect the wiki you already have. Most of this is written down already, in Notion, Confluence, or Glean. Connect the tool on the Apps tab, switch it on for a chat from the composer’s + menu under App connectors, then ask Addison to read a page or a space and turn it into guides. Nothing to copy across by hand.
  3. Write it yourself. Choose Create Manually for a full page with a title, a one-line description, Tags, and the body. Worth the time only when you already know the exact wording.
A guide titled 'Nowa Business Overview' with the description 'Provides an overview of what the business does, 2026 strategic priorities, company structure, and KPIs,' fields for Created by ('Jojo Chen'), Status ('Global'), and Tags, and a Markdown body with headings including 'Company Overview' and 'Scale & Structure.'

A guide, open for editing

Metrics

A metric is one number, defined once, with its calculation attached. Define “gross margin” as a metric and it computes the same way everywhere: in a chat answer, on a dashboard card, and in a scheduled report. A guide tells Addison what a word means. A metric decides what the number is.

What to make a metric

The test is disagreement. If two people in your company have ever reported different numbers for the same term, make it a metric. Common candidates:
  • Revenue and margin: net revenue, gross margin, contribution margin
  • Recurring revenue and retention: ARR, MRR, net revenue retention, churn rate
  • Cash and working capital: burn, runway, DSO, cash conversion
  • Volume and throughput: units shipped, orders processed, open backlog
  • Service levels: on-time delivery rate, fill rate, first-pass yield
  • Efficiency: inventory turns, cost per unit, utilization
  • Ratios where the denominator is the real argument: average order value, cost per order, margin percentage
Leave the rest as guides. A metric is a single number computed one way, not a segmentation, a policy, or a slice of a number you already have. “Net revenue for EMEA” is a filter on net_revenue, not a second metric. Check for a near-match before you add one. Near-duplicates with slightly different names are how a catalog stops being trusted. Addison checks for you before it drafts.

Create a metric

Three approaches, depending on what you already have.
  1. Ask Addison. Describe the number: “Profit is Revenue minus Cost.” Addison finds the table behind it, writes the calculation, runs it to confirm the result is plausible, and validates the definition before handing back a draft. It asks about anything ambiguous, such as the time grain or whether deleted rows count, rather than guessing. Addison can save a draft into a project; publishing to the whole organization stays a human decision.
  2. Bring an existing semantic layer. If your metrics already live in dbt, Cube, LookML, or Snowflake semantic views, attach the YAML to the same prompt instead of retyping them. Addison maps each definition onto a table it can query, validates it, and creates a whole family in one pass, including the ratios and derived metrics built on the simple ones beneath them. Two things to know:
    • Summation stores definitions as dbt semantic-layer (MetricFlow) YAML, so dbt files carry across most directly. The result is the same from any of the others.
    • Every metric needs a table already connected to Summation with a date column in it. Bring the layer across one domain at a time, because a smaller batch is one you can actually review.
  3. Write the YAML yourself. Choose Create Manually for a panel with the two things a metric needs: the Table it is defined against, and the Metric Definition YAML. Name, description, and tags are read out of that YAML rather than typed beside it.
The 'Create Metric' panel with a Table field above a 'Select table' dropdown and the hint 'Select the table this metric is defined against to save it,' then a 'Metric Definition' heading above an empty code editor showing line number 1, with Cancel and a disabled Save button along the bottom.

Creating a metric by hand: a table and a definition

Metric definition example

State the grain and the exclusions in the description. Addison reads it to decide whether this is the metric a question is about, so a description that restates the label tells it nothing.

Tags

Tags group related guides and metrics. Both kinds draw on one shared vocabulary per project, so tagging a revenue guide and a revenue metric alike keeps a definition and the rule that governs it together in the listing. For a guide, use the Tags field in the editor. For a metric, tags come out of the definition’s config.tags rather than a separate field:
The picker offers every tag already in use across the project’s guides and metrics, with a search box, so you pick from the vocabulary that exists instead of starting a parallel one. Typing a new tag creates it, and it joins the vocabulary once the guide is saved. Three conventions worth setting before the library grows:
  • Tag by domain, not by owner. finance, supply-chain, and marketing survive a reorg; janes-team doesn’t.
  • Keep the vocabulary small. A tag earns its place by being reused. One asset carrying a tag of its own organizes nothing.
  • Match case exactly. Finance and finance are two different tags. Pick one form and hold to it, because nothing merges them later.
Tags appear as a column in both listings. There’s no filter-by-tag control yet, so search is how you narrow a long list today.

Save context as you work

When a correction in chat will matter again, save it instead of re-explaining it next month. Ask Addison in that same conversation to keep it as a guide or a metric. You can also highlight a term in a report: Addison proposes a definition inline, and Add saves it.

Global vs. project

Every guide and metric lives at one of two levels:
  • Global (the Knowledge page, from the left nav): shared with the whole organization. Addison uses it in every project. Add something here when every project should have it.
  • Project (the Knowledge button at a project’s top-right): scoped to that one project.
Project Settings for 'My Project' with Tables, Skills, and Knowledge tabs, on the 'Project Knowledge' page: 'Add context that this project relies on most, and tailor any definition just for this project. Addison already uses global Knowledge.' Guides and Metrics tabs sit above an empty state: 'No project guides yet, Guides added to this project will appear here.'

A project's own Knowledge

Projects and the global library work together.
  • Add a global guide or metric to a project to prioritize it. Addison can already see everything in global from any project; adding one to a project just makes Addison reach for it first.
  • To customize a global guide or metric for one project, add it to the project and edit it there. Your edit applies only to that project. The global version doesn’t change.
  • To share a project’s guide or metric with the whole organization, click Publish to Knowledge. An admin has to approve it before it goes live everywhere.
  • Or create a guide or metric that stays in the project, with no plan to publish it.

How Addison uses Knowledge

Without Knowledge, Addison reasons from table and column names and your catalog. Knowledge replaces those guesses with your team’s answers:
  • “Revenue” resolves to net revenue after refunds, not gross
  • “Active user” means logged in within the last 30 days
  • Average order value is computed the one way your team has agreed on, not reinvented per chat
You can see which definition a number came from. Cited figures in a report are underlined, and clicking one opens a Value trace. When the figure resolved through a governed metric, the Summary names the table and the metric behind it, so the number links back to the definition rather than to an ad-hoc calculation.
A 'Value trace' dialog over a report showing the value $432,525 above Summary, Data, and Source tabs. The Summary tab lists Table 'fact_charging_session' and Metric 'total_revenue' as a link, with 'Last updated Sep 3' in the footer.

A Value trace: the number resolved through a governed metric