Skip to content
Changelog
.md

Changelog

What’s new in the Intuizi API, the MCP server, the CLI, and the console.

Notable changes, newest first, since the first generally available release. This page shows the latest month in full, and every earlier month has a page of its own.

Each change is tagged with what it affects:

  • API the REST API. The MCP server and the CLI call the API, so they get these changes too.
  • MCP the MCP server, its tools, or connecting an MCP client.
  • CLI the Intuizi CLI.
  • Console the Intuizi console.

November 2026

2026-11-19

  • MCP: the Transactions dataset has its console name MCP

    The MCP server now names the Transactions dataset the way the console does. Its audience type is Transactions, its reference group in lookup_reference is transactions, and the activation limits of create_activation and preview_activation are transactions_spend_limit, transactions_spend_min, transactions_spend_max, transactions_txn_limit, transactions_txn_min and transactions_txn_max. Tool results use the same names. An audience built from the retired merchant-level dataset reads as type TransactionsLegacy. The server still accepts the earlier names, so a client that keeps them goes on working. The REST API, the CLI and the webhooks are unchanged.

2026-11-17

  • MCP tokens and one-click connections no longer work on API v1 API MCP

    An MCP token, and the token a one-click (OAuth) connection gets, are for the MCP server. On API v1 and on GET /api/user/info they now get 403 with “An MCP token does not work on this route. Use it on the MCP endpoint, POST /api/v2/mcp, or use an API token.” The MCP server and its tools are unchanged. An integration that calls API v1 with one of them needs an API token.

2026-11-14

  • Charts: true distinct counts in Web Report, CTV Report, and CTV Insights Console

    The charts of these three data streams no longer add distinct counts up. A device or an IP used to count once for every row it appeared in, so tiles such as Unique Devices and Avg Daily Unique EIDs could read far above the size of the audience. They now count each device and each IP once, and each signal once. In Web Report, Signals per EIP is now Signals per Device, and Domains by Country is a flat table that lists the largest domains with their countries. In CTV Report and CTV Insights, Total Signals counts signals, and CTV Taxonomy by Unique EIPs is a bar chart. Audiences built before this change keep the charts they were built with. A new build shows the new numbers.

2026-11-13

  • SKUs: data coverage in the Audience Architect, devices refreshed daily Console API

    In the Audience Architect the dates of a SKUs dataset now show how far the data reaches: the newest purchase day and the newest day with devices. A warning appears when the picked dates start before the first purchase day, include days with no or few purchases, end after the settled day, when the newest days still read low, or end after the newest day with devices. The devices seen on a home network are now refreshed once a day (they were refreshed once a week), so a purchase click gets its devices sooner. See Coverage and freshness.

  • SKUs: coverage and insights reads API MCP CLI

    Get SKUs Coverage says what the SKUs data covers: the first and the newest purchase day, the settled day, the newest day with devices, and the days with no or few purchases. Get SKUs Insights ranks the product brands, the merchants, or the product categories by orders over the last 7, 30, or 90 days, with spend, the average order value, home networks, devices, and the share of orders, optionally within one category or one merchant. Rows with fewer than 10 home networks are not listed, nor rows with fewer than 10 orders. Both reads return an empty list until their first daily update has run. lookup_reference takes dataset: skus with catalog: coverage and catalog: insights.

    CLI v0.1.17 adds intuizi reference skus coverage and intuizi reference skus insights (--dimension, --period, --category, --merchant, --limit): see Catalogs.

  • SKUs: spend filters, a dataset page and a Data Overview card API MCP Console CLI

    A SKUs block takes seven optional spend filters: item_price_min and item_price_max (USD per purchased item, a filter on each purchase line), spend_min and spend_max (USD a home network spent across its matched purchase lines in the window), orders_min and orders_max (that home network’s orders), and spend_top_percent (50, 25, 10, 5 or 1: the top share of home networks by spend, ties at the cut kept). The rank behind spend_top_percent covers every home network with a matched purchase and is computed before the spend and orders bounds. A minimum above its maximum is a 422, and the keys are refused on other dataset types, except spend_min and spend_max, which AffinityTransactions already accepts with its own meaning. See Spend filters. In the Audience Architect the SKUs tab has a Spend row (Item price, Home network spend, Orders min, Top spenders). create_audience and estimate_audience_size accept the seven keys.

    The same day the console gained a SKUs dataset page and a SKUs card in Data Overview.

    CLI v0.1.16 adds --item-price-min, --item-price-max, --spend-min, --spend-max, --orders-min, --orders-max, and --top-spenders (50, 25, 10, 5, or 1) to --type skus on audiences create, audiences estimate create, and audiences quote. A bad value, a minimum above its maximum, a --top-spenders outside the set, and any of them on another type or next to --file are refused before anything is sent, exit 2. Amounts are plain decimal numbers with at most 6 decimal places. A --category given twice is sent once. See Audiences.

2026-11-12

  • Credits read the current calendar month API CLI Console

    Get Credits, the credits block of Get Usage, and Get Credits Statement now cover the current calendar month in UTC for every organization, also during a contract term, so period.type is always month. used counts the runs charged this month, net of their refunds. A refund counts with its run, in the month the run was charged, so a refund of a run from an earlier month no longer lowers this month’s used. The share used and state follow the month, and so do the console’s Credits tab and its credit alerts.

    Plan credits now arrive as one grant a month, a Plan credits line on the 1st of each month of the term. An expiry line is dated at the end of the credits’ last day, so the expiry of a month’s plan credits belongs to that month.

  • SKUs: product categories and the purchase device API MCP CLI Console

    A SKUs block takes categories: category ids from the new Get SKUs Categories read, the Google Product Taxonomy with finer levels of our own (Dry and Wet Dog Food, for example). A category selects itself and every category under it, and it narrows brands and merchants on the same order line: Dove with Deodorants is the buyers of Dove deodorant. A block now needs merchants, brands or categories. Get SKUs Brands takes categories to list the brands sold under them.

    device_types, device_oses, device_makes and device_models filter the orders by the device of the purchase click, with values from Get SKUs Device Values. The audience is still every device seen on that home network. In the Audience Architect the SKUs tab has a Categories picker (search, or Browse the tree) and a Purchased On row. lookup_reference reads both new catalogs with dataset: skus.

    CLI v0.1.15 adds --category (a category id or name) and --device-type, --device-os, --device-make, and --device-model to --type skus on audiences create, audiences estimate create, and audiences quote, and a SKUs dataset from flags now needs --merchant, --brand, or --category. reference skus categories (--parent-id, --search) and reference skus device-values (--field type|os|make|model) read the two new catalogs, and reference skus brands --categories lists the brands sold under those categories. See Audiences and Catalogs.

  • New dataset: SKUs, devices by the product brands and merchants of online purchases API MCP CLI Console

    A SKUs dataset builds an audience of the devices seen on the home networks of the purchases of a product brand, or of the purchases from a merchant, in a period. The dataset is available on request: your Account Manager enables it for your organization.

    The data is online purchases at product level from a panel of affiliate-tracked orders, and it is strongest at large United States retailers. The audience holds the devices seen on the home network each purchase click came from, so an order with no tracked click adds no device. Read what the data is before you size a campaign on it.

    In Create Audience and Estimate Audience Size, a SKUs block takes merchants or brands (or both), the purchase dates, and optionally location.countries, location.states, and max_devices_per_ip, its Match Strictness. It takes no signal_providers. An estimate is answered by a worker.

    Two new reference reads return the keys a block takes: Get SKUs Merchants and Get SKUs Brands, which you search by name. The MCP tools take the same block, and lookup_reference reads both catalogs with dataset: skus.

    A SKUs dataset is built alone or with one other dataset of the types Combining a SKUs dataset lists, and an audience with a SKUs dataset runs no analysis and no refine.

    Until SKUs has a credit price, an organization billed in credits gets a 422 with data.error dataset_not_priced on Create Audience and Create Schedule for a SKUs audience (an estimate is still answered). See Errors.

    In the Audience Manager, an organization that has the dataset picks it as a Data Set Type and sets the same filters: merchants, product brands, purchase dates, countries, states, and Match Strictness.

    CLI v0.1.14 builds a SKUs dataset from flags: --type skus with --merchant, --brand, and --match-strictness, on audiences create, audiences estimate create, and audiences quote (a SKUs quote answers 422 until SKUs has a credit price). reference skus merchants and reference skus brands read the two catalogs. See Audiences and Catalogs.

2026-11-10

  • New POI audiences and estimates now apply GPS Strict Filtering by default API MCP CLI Console

    A new POI dataset now gets two quality filters unless you choose otherwise: anomalous_devices and strict_gps. The Audience Manager uses the same default, so the same recipe gives the same size on every surface. In the Audience Manager, GPS Strict Filtering replaces Filter Anomalous POIs as a default. Filter Anomalous POIs is still there, off by default.

    strict_gps drops every visit at a spot where more than 500 devices of one signal provider were seen in one calendar month (for the current month, the newest month whose list is published). A spot is the coordinates rounded to 4 decimals (about 11 m). These are data errors, such as city-center points and grid points. Turn it off for airports and stadiums, where real crowds can stand on one spot.

    In Create Audience and Estimate Audience Size, a POI dataset that leaves filters out (or sends null) gets the default set. filters: [] turns every filter off, and a list applies exactly that list. The CLI has --no-filters for filters: [], and --filter replaces the default. The MCP tools follow the same rule. Next to an AffinityTransactions dataset no default applies, and a non-empty filters list is refused with 422. See POI filters.

    What changes for you:

    • Sizes are lower. New POI audiences and estimates that leave filters out count fewer devices than before. For example, Starbucks stores in California for October 2026 hold 102.8M devices with no filters and 3.5M with the new default. Send filters: [] to get the old API, MCP and CLI numbers, which had no filters. To match the old Audience Manager default, send ["anomalous_devices", "anomalous_pois"], or tick those two in the Audience Manager.
    • Stored recipes keep their filters. Existing audiences and schedulers, and clones of them, keep the filters they were created with. To build an older API audience again from its normalized_payload, add "filters": [] to each POI dataset whose filters is missing or null: without it, the dataset now gets the default set.
    • A POI estimate is answered at once only with the default set. An estimate that leaves filters out, or sends exactly the default set, can still come back as a fast answer. An estimate with filters: [], or with any other set, is answered by a worker: slower, and it counts toward the build budget.
    • Estimates made before this change do not match creates made after it. An estimate that left filters out before this change had no filters. After it, the same body gets the default set, so its recipe_hash differs. Estimate again before you create from an older estimate. An identical estimate sent across the deploy runs again instead of sharing the earlier run, and older estimates lose the admin “built as audience” link. normalized_payload now shows the filters that were applied.
    • Two POI datasets must resolve to the same filters. That rule is not new, but its result is: a body that leaves filters out on one POI dataset and sends filters: [] on the other was accepted before, because both meant no filters. Now the first resolves to the default set and the second to none, so the request is refused with 422. Send the same list on both, or leave it out on both.
    • Cross-visitation is unchanged. It keeps its own filters, Anomalous Devices and Anomalous POIs, and GPS Strict Filtering does not apply to it.
    • CLI. CLI v0.1.13 adds --no-filters. Older CLIs cannot opt out by flag: their help still says “none by default”. They can send a --file body with filters: [].

2026-11-04

  • An SCID file cohort can match SCID links from up to the last 3 months API MCP CLI Console

    Create Cohort takes match_months on an SCID file import: the months of SCID links to match, 1 (latest month) to 3. Each SCID uses its most recent month in that window. It defaults to 1, so a create that does not send it gets the same result as before. The MCP tool create_cohort takes it too, the CLI sends it through --file, and the console’s SCID file import offers it as Match Window.

    A completed SCID file import now also reports total_scids, the distinct SCIDs in the file, and matched_scids, how many of them have at least one device in the window, in the cohort read, the list and the cohort webhook events. The console’s Cohorts page shows them as “SCIDs matched: 8,849 of 13,051 (68%)”. Both are null for other cohorts and for SCID cohorts built before this change. See Get Cohort.

2026-11-03

  • A cohort from a file can keep only the rows with chosen column values API MCP Console

    One file can hold several audiences, for example one row per person and channel with a column naming each row’s group. The new Scan Cohort File counts the rows of a csv or gzip file by the values of 1 to 5 of its columns before the cohort is created, and Create Cohort takes the values to keep as row_filters: all entries must match, and any value of an entry matches. Without row_filters a cohort keeps every row, as before. row_filters applies to a file source only: a create from an audience that sends it is refused with 422. The MCP server adds the tools scan_cohort_file and get_cohort_file_scan, and create_cohort takes row_filters. In the console, Create Cohort From File has “Keep only rows where”, with the row count of every value and the rows kept before the cohort is created.

2026-11-02

  • The Connectivity signal providers list only providers with Connectivity data API MCP CLI Console

    Get Signal Providers with dataType Connectivity, and the Signal Provider list of a Connectivity dataset in the Audience Architect, now list only the providers that have Connectivity data in at least one built week. Before, they could list a provider without it, so selecting every provider, or leaving signal_providers to the CLI’s default, was rejected with 422.

  • The CLI builds Connectivity audiences and runs Network Switching CLI

    CLI v0.1.11 adds --type connectivity to audiences create, audiences estimate create, and audiences quote, with --network-kind, --network, --technology, and --min-days. audiences create takes --switch-from and --switch-to (with --switch-network and --switch-min-days) to run Network Switching on one Connectivity dataset, and audiences show prints its results. From CLI v0.1.12 the switching flags need --type connectivity. intuizi reference connectivity networks lists the networks a Connectivity dataset can name. See Audiences and Catalogs.

  • New Network Switching analysis compares the home network of a Connectivity audience’s devices in two months API MCP Console

    An audience with one Connectivity dataset can now run Network Switching, on the home internet network or the mobile carrier. Each month is read as its last full Monday-to-Sunday week: September 2026 is September 21 to 27, and October 2026 is October 19 to 25. For each device the analysis finds its home network in each week. It then counts the devices that kept their network, changed it, and switched, in total, per network, and per pair of networks. Every count is a count of devices.

    Send it as the optional top-level networkswitching object of Create Audience, with from_month, to_month, and optionally network_type and min_home_days. Both weeks must be built for the dataset’s signal providers, and a month whose week is not built is rejected with 422, which names the weeks that are built. The dataset keeps its own dates. Any other audience is rejected with 422. Connectivity is available on request, and the analysis requires additional permissions which need to be approved by your Account Manager: a request that names the block without them is refused with 403. A body that also carries a crosspurchase block is refused with 422. A schedule of an audience built with the analysis cannot be created yet and is rejected with 422.

    Read the results in analysis_results of Get Audience and in the data of the audience.completed webhook. List Audiences leaves the field out, because the results of one audience can be large. On the reads that carry it, analysis_results is always present: an empty object for an audience built without the analysis, { "network_switching": null } until the analysis finishes, and the results after. The results carry version 2 and source connectivity, and from and to hold the two weeks.

    The MCP tool create_audience takes the block, get_audience returns the results in a short view, and list_audiences leaves them out. The short view has no top_from and top_to on the networks and lists the 20 largest flows. Get Audience and the CSV file hold the full lists. The estimate_audience_size description says that an estimate cannot count devices that switched their network, and sends such a question to create_audience with the block on one Connectivity dataset. In the console, the Audience Architect offers the analysis on a Connectivity dataset with only the months whose week is built, and the Audience Manager shows the results on the audience’s Analyses tab, with the week it read under each month and an export of every pair of networks as a CSV file.

2026-11-01

  • New dataset: Connectivity, devices by home internet network or mobile carrier API MCP Console

    A Connectivity dataset targets devices by the home internet network or the mobile carrier they used on most days of a week. Its data is weekly, so the dates are widened to whole Monday-to-Sunday weeks and then moved inside the weeks that were built. The dataset is available on request, and its fields are on Create Audience and its networks are on Get Connectivity Networks.

Earlier months