Skip to main content
Addison is the AI analyst behind everything Summation does with your data. Ask a question in plain English and Addison writes the SQL, reads the results, and answers with tables, charts, and explanations — or builds a report, deck, or dashboard when the answer deserves one. Every chat is scoped to a project, so Addison works from that project’s data, files, and knowledge.
An Addison chat showing the user message '/create-report Create a Report for Olympics Players and Country Review' followed by Addison's narration interleaved with collapsed tool steps — 'Built Olympics report', 'Now let me build the tables and charts', 'Built Olympics report tables', 'Built Olympics report with 3 data tables', 'Built Olympics growth chart', 'Updated report stats', and 'Report compiled cleanly. Running verification now.' — ending in a Verification card reading 90 / 91 passed with Traces 41/41, Citations 25/25, Claims 18/19, and Queries 6/6, above a collapsed '1 finding' row.

Addison working through a request, step by step

​
Start a chat

Open Addison with the sparkle button at the top of the left sidebar, or type straight into the composer on Home. Pick the project the chat should use — that choice sets what Addison can see, and it’s fixed once the conversation starts. Without a pick, the chat uses your default project.
The top of the left nav sidebar showing the workspace name 'The Playground' with a panel-toggle icon and an orange sparkle button that opens the Addison pane, above the Search, Home, Reports, Decks, Dashboards, Tables, Playbooks, Schedules, and Applications nav links.

Toggle the Addison pane

The composer's project picker open — a searchable dropdown listing 'My Project' with a lock icon and a checkmark marking it current, followed by other projects including 'Summation Product Knowledge' and several Eval/playground projects.

Pick the project a chat uses

​
The composer

Ask in plain English — “What drove the revenue change in Q3?” beats a keyword search. The controls around the input add context and change how Addison works:
Addison's composer with the placeholder 'What can I help with today?', a + button and a project-picker button on the left, and on the right a 'Default High' model and effort selector, a microphone button showing a 'Stop dictation' tooltip while active, and a send arrow.

The composer

​
Give Addison context

Addison already reads the project’s data and knowledge. Add more for a specific question:
  • Upload Files — attach a file from your computer to this conversation.
  • Add files from project — pull in something already in the project.
  • App connectors — let Addison work with a connected app such as Shopify or PostHog. Connecting an app isn’t enough on its own: the submenu (gated by the ui_mcp_connectors flag) lists only active connectors, and each carries a per-chat toggle you must switch on before Addison can use it in this chat. If none are connected, Add new connector takes you to Connectors.
  • @ mentions — the fastest path: type @ and pick a file or folder, each shown with its path.
The composer's + menu open with Upload Files, Add files from project, and App connectors; the App connectors submenu shows Gmail, Google Drive, and Slack icons above 'No connected apps — Connect an app to use in chat.' and an 'Add new connector' button.

Attach files or reach an app

The @ mention picker open over the composer, headed 'My Project' and listing folders and files with their paths — Reports, Table Catalog (/Data Context), Untitled Folder, activated_buyer_app_installs_z…, applicable_roles.catalog (/Context/Data Catalog), buyer_activation_status_zoq.ca…, Buyer Engagement Overview.sdash (/Dashboards), buyer_touches_by_type_zoq.ca…, community_engagement_rankin…, customer.catalog, Customer Insights Report.smd, lasso.png, Launch Week Orders.xlsx, and machine_learning_files.ca… — with '@' typed in the composer and a Default / High model selector below.

Mention a file with @

​
Skills

Type / for skills — named workflows that put Addison straight into a specific job instead of describing it. /report writes a report, /verify-and-correct fact-checks work against its sources, and /remember teaches the project something durable.You rarely need them — asking in plain English gets you to the same place — but they’re the shortest path when you know exactly what you want.
The Skills menu open above the composer after typing a slash, listing /charts-editor, /context, /create-demo-data, /create-report, /incorporate-feedback, /playbooks, /remember, /report (highlighted), /summation-sync, /table-editor, and /verify-and-correct.

The skills menu

​
Choose a model

The Default model suits most work. When your workspace allows per-request model selection, the selector lists the models you can pick (grouped by provider, which may include Anthropic, OpenAI, DeepSeek, xAI, Z.ai, and Kimi) and sets Effort, which trades speed for depth of reasoning (this control is labelled Mode for models that expose mode options). Raise it for a gnarly analysis; leave it alone for a quick lookup. When per-request selection is off, only the default model is returned and there’s no provider list to open.
The model selector open above the composer, listing providers Anthropic, OpenAI, DeepSeek, xAI, Z.ai, and Kimi each with a submenu arrow, and an Effort row set to High at the bottom. The composer's trigger below reads 'Default High'.

Model and effort

​
Read a response

Addison narrates as it works — each step it takes (checking a skill, finding tables, running queries) appears as a collapsed line you can expand to see exactly what it did. That trail is the point: you can audit how an answer was reached, not just read the conclusion.Answers come back as prose, tables, charts, or single metrics, depending on the question. When Addison produces an artifact, it lands as a card at the end of the response with an Open button — and it’s a real file in the project from that moment on. Copy the response, retry it, or rate it with the controls beneath.
The end of an Addison response describing a built dashboard: after narration ending 'Dashboard is built and validates cleanly', it summarizes 'Olympics Players and Country Review' as two tabs filtered by Season (Summer / Winter) — a Countries tab with 4 KPIs (total medals, competing NOCs, unique athletes, Games held), a Top 15 countries bar chart, a Gold/Silver/Bronze/Total medal breakdown table, a 5-country trend line since 1896, and Top 10 sports by medals; and an Athletes tab with 2 KPIs (Female share 25.1%, Medalists 28,251), a Top 20 all-time medalists table, an age-bucket column chart, and a male-vs-female participation line, plus a data note of 271,116 rows across 1896–2016. Below it a card reads 'Olympics Players and Country Review.sdash' with an Open button, above copy, retry, and thumbs up/down controls.

A finished response, with the artifact it produced

​
Chats and layout

Every conversation is its own thread, kept under Recent chats in the left sidebar. Threads hold context, so follow-ups work — “now break that down by region” needs no repetition of the original question.The pane itself adapts: switch between Sidebar (alongside your work), Floating, and Full screen from the panel icon in the pane header. Sidebar is the default and keeps Addison next to the artifact you’re editing.
The full left sidebar with nav links (Search, Home, Reports, Decks, Dashboards, Tables, Playbooks, Schedules, Workflows, Applications) above a 'Recent chats' section listing threads such as 'Olympics Medals Leaderb…', 'Japan Olympic Champions', 'Olympics Athletics Guide', and 'US Olympic Champions', with Projects, Knowledge, Connectors, and Dagster jobs pinned at the bottom.

Recent chats

The Addison pane header menu open, offering Sidebar (checked), Floating, and Full screen, beside a + button for a new chat and a close button.

Sidebar, floating, or full screen

​
Get better answers

  • Add knowledge — metric definitions, glossaries, and domain context make Addison’s answers match how your business actually measures things.
  • Enrich your catalog — descriptions on tables and columns help Addison pick the right data without being told.
  • Be specific — “Show monthly revenue for Q1 2026 by region” beats “show me revenue.”
  • Keep the thread going — follow-ups inherit everything above them, so refine rather than restart.
  • Check the steps — when an answer surprises you, expand the tool lines to see which tables and queries produced it.