Everything in the app

Overscore keeps a company's data in one place and answers from it. Your store, your books, your database, your spreadsheets and your ad accounts are connected once and stay in step on their own. Then you, your team and your AI agents ask questions, build dashboards, agree what each number means, and send results to the systems you work in. Every number carries a line that says where it came from, written by the software, never by the AI.

This page is the map: every screen in the order it sits down the left side, what you can do on it, and the page that has the detail.

Down the left side

Home sits alone at the top. Dashboards, Chats and Inbox are for everyone. Under Data, Warehouse, Models, Sources, Destinations and Context are for whoever builds. Playbooks sits at the bottom, above the line over Members and Settings, with you under them.

• Your organization, at the top: its logo, its name and its plan. Click it to switch to another organization you belong to, start a new one, or open People and Plan and billing. Each one under Switch to is its logo (or its initials on its colour) and its name; the one you are in is greyed, and one still being built says so under its name. One sign-in can belong to many organizations, and the whole app follows the one you choose. An admin uploads the logo in Settings, Company: Upload a logo.
• Ask, under it: a chat beside the page you are on, the same chats as on Chats. The same Ask is in the bottom right-hand corner of every page. See Ask a question.
• You, at the bottom: your photo, your name and your role. Click it for Your sign-in (your photo, your name, and the emails and Google accounts you sign in with), Appearance (dark, light, or the same as your computer), Your keys and laptops (on Settings, Account) and Sign out.
• ⌘K (Ctrl+K on Windows), on any screen: type to jump to a section, a dashboard, a monthly report or an analysis, a metric or a table, or to start a new chat, a quick answer or a new source. It lists only what you can open. ⌘J (Ctrl+J) opens the chat beside the page, and closes it.
• In Warehouse the sidebar folds to its icons, with a thin line where the Data heading was, so the tables have the room; point at it to see the names, or open it again. It folds the same way while the chat is open beside a page.

Home

A greeting, a line that says what needs you, the idea in one sentence ("It files what it notices. Nothing counts until you say yes."), and a box to ask anything: it starts a chat. On a new company that is all, with Where to start: four doors, talk about the business, connect a source, open a dashboard (the first page published, or, with none yet, build one with the chat) and take the tour, first for someone with no other company. Each door leaves once it is done (the business on record, a source connected, a published page opened in this browser, the tour taken), and the heading with the last. Connecting Claude Code is on Settings, Agents. The rest shows once it has something: Sources (how many, and any that broke), Context (how many things it knows, by area), Accuracy (an admin's, once a test exists: how many of the numbers you already know came back right last night), AI this month (the company's AI against its limit, once any is spent), Waiting for a yes (the first five things in the Inbox that ask you something, each a row as the Inbox draws it, opening its card) and Recent (the latest syncs, builds and checks).

Dashboards

Three tabs: Dashboards, Alerts and Emails. See Dashboards, Alerts and Emails.

• Dashboards lists every dashboard and report, as cards or a table, with folders, stars, a search and a sort. Its own tabs are All, Dashboards and Reports; Reports holds the monthly reports and the analyses. New dashboard asks for a name and opens the builder: a chat beside the page that builds it, looks at it, and fixes what it finds.
• A dashboard is the page itself, live, in your company's colours. Click a number for what it is, in a sentence, with Ask about this number and Show the rows (and Prove it for whoever builds, and for a group given Proof). Under its title, a link when something needs doing ("4 numbers aren't metrics yet"). Its tabs: Overview (the page), Where the numbers come from (the one thing to do, Define them or Confirm it, then each card in its own words with one sentence of what it reads, every name a link; a row opens its card on the page) and Settings (its look, its versions and Real time). Each save is a version, and publishing picks the one viewers see. Download saves the page as a PDF, or each table on it as CSV or Excel. Full screen shows the page alone.
• Alerts: each one watches one number and tells you once when it crosses its line, with a card in the Inbox and an email.
• Emails: the numbers that matter, sent on the day and at the hour you choose. Each is made in a chat and goes out only once a person turns it on.

Chats

One assistant, as many conversations as you like, newest first, each with a name you can change, and a search over all of them. A new chat starts with four ways in and the box; the box on Home starts one too, with your question. See Ask a question.

• It answers as you: it reads only what you may read, and every figure carries a line that says how far to trust it.
• When the answer is a split, a ranking or a trend, it draws a table or a chart. Under it, Put this on a dashboard, and Keep or share: Save as an analysis, Download as Excel or CSV, Email it, and Where from.
• It builds or changes something (a dashboard, a metric, a model, your fields on a table) only when you ask, and nothing leaves Overscore until you press Send on its card. Reading another website asks you first; a search of the web shows its questions on a card and waits for Run it.
• Ask someone to check, under an answer, sends it to a teammate. Right and Wrong under an answer teach it.

Inbox

What needs a person, one row per thing. The number beside Inbox is how many rows wait. See The inbox.

• Waiting on you: what it heard and needs a yes, what broke (with the fix, where it has one), numbers and finished months that moved, metrics nobody has confirmed, checks teammates sent you, and figures on a page that nothing explains. Click a row for its panel and its buttons; Later sets it aside.
• Later: improvements noted and not made yet, grouped by what they are about, with Work on it in a chat and Mark it done.
• Handled for you: what closed itself this month, and what happened to it.

Warehouse

The first row under Data: three panes, the way a database tool lays them out. See The warehouse, Your fields on any table and Tables you fill in yourself.

• On the left, every table by dataset, with its rows; views among the tables, and tables people fill in marked with a grid and a pencil. Find a table narrows the list.
• In the middle, the table you pick, rows first: its rows fill the page, and beside Rows are Columns (each column on one line, with a drawing of what it holds and what it means; Edit beside a meaning writes it in place, Return saves, and a meaning the AI guessed says not confirmed yet until a person says Yes or corrects it; a column opens its own panel) and History. About this table opens its card, what reads it, the notes people left, your fields and where it comes from in the pane at the right. The card has one line only the AI reads (Only the AI reads: a join key, a unit, a trap, like "amounts are in cents; join to shopify.orders on orders_id"); it is never shown to a viewer, never drafted by the AI, and the chat reads it before the columns. Query opens it in the editor; Download saves every row.
• At the top of the list on the left: Files (every file anyone gave the business: a document, a spreadsheet that became a table, a Google Sheet that refreshes). Above the middle: your open queries, Lineage (what feeds what), and + for a new query. A query can be written for you: ask in words.

Models

SQL kept under a name, as a view, or as a table built after what it reads lands, with tests that must find nothing. Four tabs. See Models.

• Models: each model's state in a word with its time, its tests, its last seven runs and when it refreshes; a model's page with its SQL, what it reads, what reads it and where its table goes, its columns and its tests. Map: every source, table, model, saved query, metric, dashboard and destination, opened on the models, and how the data flows between them. Runs: every build, what it built and who started it; every sync, check and send behind Show every run. Jobs: what builds on a schedule, its last five runs and when it runs next.

Sources

Every connected system, with its status, its last sync and its schedule. Add a source shows the systems Overscore connects to, each with a guide beside its form; a webhook address; your own Postgres; a source built from an app's API docs. A source's page has its syncs, its tables and its settings: Replace the key, Re-sync from the start, Pause and Disconnect. At the top, Upload a file and Add a Google Sheet are two big cards of their own; a Google Sheet you add is listed with the systems, and every file is kept under Warehouse, Files. See Connect a source.

Destinations

Where your data goes: Meta, Google Ads, HubSpot, your own BigQuery, a bucket of yours, or a web address of yours. Three tabs. Sending lists every send, with what it sends, where to and how it last went; Places lists the places, with Add a place; Sent is every row that left. Send something makes a send, and shows the next row exactly as it will leave before anything goes. A destination is on the map at the right edge, after the table it sends, and on that model's page under Read by. See Sending data out.

Context

What Overscore knows about the business, which every answer reads first. Three tabs. See Context and definitions.

• Library: folders for the company, its people, customers, products and systems, the metrics, the rules, the timeline, notes and checked answers. Click a row for where it came from, every version, and the pencil that edits a line.
• Talk it through: the interview, one question at a time; it never asks what it already knows. Tell it something: say a fact in a sentence.
• Every folder shows, even one with nothing in it yet: it says none yet, and what would fill it. The accuracy tests, questions whose answers you already know, asked every night to test the AI, are under Settings, Company, as Accuracy.

Playbooks

At the bottom of the left side, above Members: worked paths through the app for one job people have, like a report that arrives by email or lead scoring sent back to HubSpot. Each step opens the screen where it is done, and an AI step shows the prompt it uses. Words and links, the same for every company. See Playbooks.

Members

At the bottom of the left side, above Settings: everyone in the company and their groups. Two tabs.

• People: everyone in the company, their role (Admin, Builder, Viewer, or "Viewer · Reps" for a viewer narrowed to a group) and when they were last in. Before a first visit, their row shows the current invitation status. Add a person sends an invitation. Open a pending person to resend their invitation without changing their group. See as, on a person's panel, shows the app as they get it, reading only.
• Groups (an admin's): what each group may see. A viewer in no group of yours sees every published dashboard; one in a group sees only the dashboards, folders and metrics shared with it, and where a number comes from only when the group has Proof.

Settings

• Plan and billing: your plan in one line and what it includes (its sources, models, storage, how it is billed and its AI), the AI used this month against your limit (we set it with you; sorting what you add to the library is on us), one line under the plan saying what Overscore sets (the plan, the AI limit and the machine) and who to ask, and today's meters. On Free, the month's limits and Use this for my company. An admin clicks Your AI this month for AI this month, by who asked for it: each person with how many chats and what they cost, most first, then the AI on its own apart (the watcher, emails, Python models, alert cards) with what each ran, and a month picker; sorting what you add to the library is one line, on us. A person's row opens their chats, each with what it cost and a link to its record; a runner's row opens the runs, emails, models or alerts it read. See Limits and plans.
• Company: the company's settings, one row each under four headings. The company: its name (an admin changes it there: what people see in the app and in its mail, while its address stays the same), the logo, the time zone, the brand, and who gets the morning email. The AI: how much it asks you, the AI switch, AI on its own (the AI that acts without being asked, which reads failed runs, looks into alerts, writes email summaries and answers a Python model's asks up to a ceiling a month that we set, shown on Plan and billing under the AI this month; off, none of it runs and the chat goes on), web search (Ask each time, On or Off, with today's searches), the sites the AI may read without asking, keys for outside agents, who can confirm a metric, what every chat starts with, the house rules (a page of their own, as Accuracy is) and Accuracy. Your data: behind the scenes, live links, Download as a dbt project and Take everything (the whole account as one zip, read-only: Take everything with you), and the company's address with its MCP address, the same ones Agents shows. Last, the Danger zone: Delete this company opens a panel where an admin types the company's name and presses Delete in seven days. A switch or a choice saves as you change it; a row with a value opens a panel on the right, which saves with Save. A member sees the values, without the controls.
• Alerts: who gets told, and about what (How Overscore tells you): every member, with what they get in a few words (Everything, Everything except 1, an alert's name and 1 more, Nothing) and how (Email, phone). Click an admin for a switch per dashboard, one for all sources and one for each data problem, saved as you flip it. Click anyone else for what they were added to, each a link to it. Under them, Also posted to: each Slack channel with the same switches, and Add a Slack channel, where an admin pastes a webhook and tests it.
• Agents: connect your Claude. Connect your Claude has the lines that set up Claude Code, the MCP address for any other agent, and Computers signed in, each laptop logged in with ovs login. Make a key chooses which datasets a key may use and at what level; once your company has one, the Keys table lists them, each with Revoke, the bin at the end of its row. Overscore keeps one key of its own for its checks and updates: one line says so, and Show it lists it. What agents did is in Activity, under Outside a chat. See Agents and keys and Use with Claude Code.
• Account: your email, the appearance, which Google you connected (to read the Google Sheets you add; it connects when you add a private sheet, on Sources) with Disconnect, notifications on this phone, Draw tables and charts in answers, and Sign out; under it, Your keys and laptops: the computers signed in as you, each with Log out, and your own key.
• Activity: every conversation, turn by turn, with what it looked up, ran, read and made; Usage (what gets used) and Questions (what people ask); Export saves the record of what was done in the company's name. Everyone sees their own; an admin sees everyone's.

On a phone

A bar along the bottom: Home, Dashboards, Chats, Inbox and More. More holds your organization, Context, Playbooks, Members, Settings, you, and notifications on this phone. Building (the warehouse, models, sources and destinations) is left to a bigger screen. A viewer has four: Home, Dashboards, Chats and More. See On your phone.

What someone in a group sees

A viewer, a client looking at the dashboards, has three screens, Home, Dashboards and Chats, and their own account under Settings. Home is the pages they starred, as cards, then the chats they can pick up; with nothing starred yet, every page they may open. Dashboards is their dashboards, starred first, then each folder, then the rest, each card with its picture and one line: its headline and when its numbers are from. A page has one bar: where they are, when its numbers are from, Download and Full screen. A number opens as one sentence of what it is, with Ask about this number and Show the rows. Ask, above Dashboards and in the corner of every page, opens their chat beside the page they are on; it answers from what is shared with their group, and their alerts and emails are under Settings, Account. See Dashboards, Who sees what.

On Free

Free, for learning: a machine of your own that sleeps while you are away, your own files and Google Sheets, up to three connected systems, and your own Claude Code working through ovs and MCP. Its limits are 100 tables, 1,000,000 rows as you see them and 1 GB of data in all; 10 awake hours a month; three connected systems, each pulled once a day. Overscore's AI sorts what you add to the library, on us. The chat (past a new company's three welcome messages, on us), quick answers, the builder and the interview are your own Claude's there, and nothing else runs on a timer. See Limits and plans.

What runs on its own

• Each source syncs on its schedule (once a day on Free), and a model that reads a table is built again after it lands.
• Every few minutes each source and model is checked against its schedule. A failure is tried again, or becomes one row in the Inbox with what went wrong and the fix.
• After new rows land, the numbers on the dashboards opened in the last two weeks, on the alerts that are on and in the scheduled emails are worked out again, so the next look is quick.
• Every night: each metric is run again and each finished month measured again; a guess an answer has used, or one nothing has used in a year, is asked about once (Still right?); the accuracy tests are asked; pages viewed that day are checked for figures gone stale; the day's chats are read again for what should be remembered; and everything is backed up to a second provider.

For agents and scripts

Your agents get the same doors as the app, with the same key and the same rules: the MCP address for Claude Code, Cursor and the like (Agents and keys), ovs at the command line (CLI reference) and HTTP (API reference). ovs init writes the rules and the skills an agent reads.

Where the record is kept

Every query, every step of every chat and agent, and every AI call with its cost; every number sent to a page, with its value and the version it came from; every version of a page, a saved query, a metric, a note and a model; every file opened, downloaded or deleted; and every admin action, with who did it and the key it came in with. Admin actions are kept 90 days; the rest is kept for now. Settings, Activity shows it, and an admin exports it with Export there, ovs actions or GET /actions.