Updates

Product updates & changelog

Every major release, newest first, with the ship date. Capability claims here match what's actually in the product — anything built but reachable by nobody isn't listed at all, and anything still limited to a subset of organizations is labelled Preview rather than described as generally available.

  1. Automation

    Projects: an agent plans, runs and oversees a campaign over weeks — with your KPIs, budget and approvals

    A project is a goal over a period, with accounts, an optional budget, campaigns and KPIs. The agent writes an editable plan, you activate it, and an overseer agent checks every channel's numbers and proposes or makes the changes — by default each one waits for your approval.

  2. Automation

    Agent Loops: an agent that decides when to wake up next, inside limits you set

    A Loops tab on the Schedules page: an objective, stop conditions and limits on iterations, runtime, credits and the wait between wakes. After each run the agent decides when to wake next or that it is done.

  3. Governance

    Forbidden Rules: hard limits on what an agent may read, change or delete — by platform, account, tool, who and data

    A workspace's own deny rules, checked before every call on every lane: in-app tasks, autonomous runs, schedules, alerts, dashboards, MCP, REST, raw API requests and approved queued writes. Rules are saved off and turned on from the list.

  4. Integrations

    Automatic account refresh: new ad accounts, properties and sites appear without reconnecting

    Each integration card has Refresh now and an optional automatic interval, on 26 platforms. An automatic refresh only adds and updates accounts; it never removes one.

  5. Reporting

    Interactive views in answers: KPI cards, charts, tables and editable drafts beside the report

    The agent can answer with an interactive view — KPI cards, charts, sortable tables with CSV, ad previews, forms, follow-up and re-query buttons, and editable cells that stay a draft until you submit them.

  6. Models

    Effort levels: Light, Medium, High and Extra on every task

    An Effort chip in the composer trades speed and cost for depth: more reasoning, an always-on plan critic, more repairs and longer reports at High and Extra; a cheaper, faster run at Light. Medium is unchanged.

  7. Platform

    X Ads joins as a full platform — on X's own Ads MCP, because the REST Ads API answered 403

    9 reads and 8 approval-gated writes for X Ads: accounts, funding, campaigns, line items, promoted posts, targeting and stats, plus status, line-item budgets, bids, paused creates and targeting. Campaigns are never deleted — pause is the stop.

  8. Integrations

    GitHub joins: release notes, issue triage, repository traffic — and code changes that arrive as a draft pull request

    15 reads and 7 approval-gated writes for GitHub: repositories, branches, the file tree, issues, pull requests, commits, releases, search, traffic and workflow runs, plus issues, comments, draft releases, branches, file changes on a branch and draft pull requests. Nothing deletes, nothing merges.

  9. Models

    Credit costs, measured: what a task really costs on every model and for every kind of task

    337 production tasks measured: a median of 349 credits, 21 to 8,872 credits for the same task depending on the model, and input — not output — as most of an agent's model bill.

  10. Models

    GPT-6 Luna is the default model for new workspaces, research agent included

    New workspaces plan and execute on GPT-6 Luna ($0.10 / $0.50 per million tokens). Workspaces still on the untouched old defaults were moved; anyone who had chosen a model was left alone.

  11. Platform

    Google Ads account health: suspended or not, advertiser verification deadlines, billing and policy in one read

    One read answers whether a Google Ads account can serve: its status, whether advertiser verification is required and by when, billing, disapproved ads by policy topic, and a manager's suspended children. Also over MCP.

  12. Automation

    The ad creative studio: find what is winning, recreate it as a video, review it, upload it to every platform

    Spy on competitors' ads across Meta, TikTok and Google, recreate one as a video ad from your own material with music, have it reviewed by a model that watches and listens, and upload a file that fits each platform's rules.

  13. Governance

    Decisions the agent now asks during a run: a budget the platform refuses, an account it cannot pin down, which ad to recreate

    Three decisions are a person's even under auto-approve: a money value a platform refuses, an account the run cannot identify, and which spied ad to recreate. The agent pauses, asks, and re-plans from the answer.

  14. Governance

    Permanent deletes now wait for a person on every platform, whatever the approval mode

    Removing a Google Ads campaign, deleting a Meta, TikTok or DV360 object, archiving in Google Analytics and deleting on Microsoft Advertising, Ad Manager, AdSense and AdMob now always wait for a person — auto-approve, batches and API keys included.

  15. Integrations

    The MCP server lists its tools on demand: from 171,665 input tokens a turn to about 4,800, with the same choices

    A fully connected MCP key's tool list cost 171,665 input tokens on every request. Listing tools on demand — search, describe, call — brings it to about 4,800 with the same tool choices, measured on three models.

  16. Platform

    LinkedIn Ads joins as a full platform, and the political-intent field that blocks every create without it

    25 reads and 41 approval-gated writes across campaigns, creatives, audiences and conversions. Money is already in major units, the refresh token expires on a fixed 365-day clock, and three tool groups answer "pending LinkedIn review" rather than failing.

  17. Platform

    Snapchat Ads joins, and every delete is permanent so every delete waits for a person

    22 reads and 25 approval-gated writes across three Snap API hosts. Creates land paused, conversion events validate by default, and amounts are major units everywhere because Snap's API is micro-currency.

  18. Platform

    HubSpot joins as a CRM the agent can read and write, with contacts held behind a rule rather than a setting

    17 reads and 31 approval-gated writes across contacts, companies, deals, lists, campaigns and landing pages. Creates never activate, every write needs an opt-in scope, and personal data is reachable only on a task a person started that asks for it.

  19. Platform

    Klaviyo joins: email and SMS performance beside the ad spend that produced it

    13 reads and 40 approval-gated writes across lists, segments, templates, draft campaigns and the catalog. Writes need an opt-in scope, a flow set live waits for a person, and report rates are fractions rather than percentages.

  20. Platform

    Pinterest Ads joins, with a scope list validated against Pinterest's own spec

    26 reads and 32 approval-gated writes across campaigns, ad groups, ads, audiences, keywords and the Conversions API. Micro-currency figures carry a major-unit twin, and a budget-type flip is refused unless it is confirmed.

  21. Platform

    Reddit Ads joins, and a delete needs an archive and a three-hour wait first

    29 reads and 31 approval-gated writes across campaigns, ad groups, ads, audiences and the Conversions API, plus community search and keyword suggestions. Microcurrency is converted once, and archive and delete both wait for a person.

  22. Platform

    A Stripe connector that cannot move money — and the scope argument that got it shipped

    Seven read tools that check reported ROAS against the charges that actually succeeded. Built, and held: Stripe does not currently let a Connect platform request read-only, so authorization is refused rather than widened.

  23. Platform

    Microsoft Advertising joins as a full ad platform — read it, build it, change it

    Bing Ads API v13 with its own Microsoft sign-in: 15 reads and 17 approval-gated writes, campaigns to keywords to offline conversions. Released to every organization on 21 September.

  24. Integrations

    Search-result and SEO research, built in — nothing to connect, billed at cost

    Live SERPs, keyword volume without a Google Ads account, competitor rankings, backlinks, page audits and brand mentions. Available to every organization on 21 September, with no connection step.

  25. Platform

    Search Ads 360 joins as a reporting platform — with offline conversion upload

    SA360 reporting across engine accounts, plus the one write its API actually carries. Released to every organization on 18 September.

  26. Platform

    AdSense and AdMob get write tools — and a correction about why they didn't have any

    Ad units and custom channels in AdSense; apps, ad units and mediation groups in AdMob. Released to every organization on 18 September, with one Google-side limit worth knowing before you plan around it.

  27. Platform

    Merchant Center can write — product inserts, data sources and feed pushes

    The Shopping integration was read-only. It now inserts and removes products, creates and fetches data sources, and pushes a feed — with a safety fence drawn on what a shopper actually sees.

  28. Platform

    Search Console gets its first writes — submit and delete sitemaps

    The integration could see a stale or erroring sitemap since June and could not resubmit it. Both writes now exist, and the scope that makes them work is requested by default.

  29. Integrations

    Data Manager uploads offline conversions straight from a query — and became a product you can find

    Offline conversion events now upload from a SELECT's own rows, with no file in between. Data Manager also stopped being an 11-pixel link under a disconnect button.

  30. Integrations

    Run your ad accounts from Claude or Codex — we broker the permissions and keep the audit trail

    Add Agent Planners as an MCP server in Claude Desktop, Claude Code or Codex. Your agent gets scoped access to the ad and analytics accounts your organization has connected — and every write lands in your audit log.

  31. Platform

    Google Tag Manager — read containers, draft tags, and publish only with a human review

    Ask what's firing on your site and on what trigger, then have the agent draft triggers, variables and tags. Publishing a container is never unattended.

  32. Platform

    YouTube analytics — channel reporting, anomaly detection and alerts

    Channel and video performance the agent can query, monitor and alert on, in the same run as your ad platforms — so organic video sits next to paid media instead of in a separate tab.

  33. Platform

    The first BigQuery write tool — a dedicated, approval-gated path

    The agent could always read BigQuery. It can now write to it — through a separate tool carrying the approval gate, write budget and audit trail, never through the analytics query path.

  34. Reporting

    Reports get analysis as a contract — and are written by your strongest configured model

    A report that lists numbers isn't analysis. Every report now owes five things, it's drafted by the stronger of your configured models, and period comparisons are computed rather than narrated.

  35. Automation

    The agent repairs rejected columns, scope refusals and bad tool names instead of giving up

    A refused metric, a permissions error moments after re-authorizing, or a malformed tool name used to end a run. Each is now recovered at the point it happens.

  36. Governance

    Connected accounts belong to whoever authorized them — a teammate's login no longer deletes yours

    When two people on one team connected the same platform, the second authorization removed the first person's accounts. Every integration now records who connected each account and manages them separately.

  37. Platform

    Google Analytics grants write access in a single authorization — re-authorizing no longer downgrades you

    Connecting or re-authorizing Google Analytics used to request read-only, with writes behind a separate link — so every ordinary re-authorize silently removed write access. Both scopes are now requested together.

  38. Automation

    DV360 and CM360 get threshold alerts — and the timeout that kept them out is fixed

    Alert rules now cover DV360 and Campaign Manager 360. The reason they were excluded was a property of the alert loop, not of those platforms, so the loop was fixed rather than the platforms special-cased.

  39. Platform

    Google Ad Manager writes work on the autonomous lane — with an argument-shaped safety fence

    Ad Manager's three write tools could only be reached by the planning lane. An autonomous run could restructure a Google Ads account but could not rename an ad unit.

  40. Platform

    SHOPLINE joins the autonomous write lane — and connecting a second store stops deleting the first

    Store writes existed but only on the planning lane. Worse: connecting a second SHOPLINE or Shopify store silently removed the first, and the agent could then have written to the wrong shop.

  41. Automation

    Faster runs — a two-tier read cache, merged step gates, and agent health metrics

    Most of a task's wall-clock was sitting between steps rather than doing work. A durable read cache, four sequential hops merged into one, and an /insights panel that makes the timing visible.

  42. Automation

    Agent Mail gets HTML replies, CSV attachments and per-organization ingest

    Replies now render as formatted HTML with data attached as CSV, mail routes per organization, and unknown-sender replies are rate-bounded.

  43. Governance

    Plan-patch telemetry, a sixth critic rule, and better dashboard delivery

    When the agent rewrites its own plan mid-run, the outcome is now recorded — so a rewrite that failed is visible rather than silent.

  44. Automation

    Threshold alerts reach commerce — Shopify and SHOPLINE join, 8 platforms to 10

    Orders, revenue and average order value can now drive alert rules, measured per closed shop-local day with a stopped store correctly reported as zero rather than as missing data.

  45. Reporting

    A units contract on every tool result — micros, ratios and RPM stop being guesswork

    Ad platforms return money in micros, nanos or decimals and rates as either 0-1 or percentages. Every tool result now carries an explicit statement of which, rather than leaving it to be inferred.

  46. Platform

    SHOPLINE integration — reads plus approval-gated writes

    SHOPLINE (International) joins the natively connected platforms: catalog, orders and customer reads, with store changes gated behind the same approval step as every other write.

  47. Governance

    Batch approval for scheduled tasks — one review, every write

    Scheduled tasks that propose many changes can now be reviewed as one batch instead of one approval at a time, with an age bound on the data each money write was based on.

  48. Integrations

    Native Google Sheets export — no third-party connector required

    Task output can be exported straight to a real Google Sheet using a narrow Drive scope, rather than routing through an external automation connector.

  49. Reporting

    Dashboards — shareable reporting pages that refresh themselves

    Seven built-in platform templates plus agent-designed custom pages, each on a permanent revocable link, refreshing on a schedule without a model in the loop.

  50. Automation

    Alerts and anomaly detection across eight platforms

    Threshold rules ("CPA above X for N days"), percent-change conditions and statistical anomaly detection, running on a cron and emailing when something moves.

  51. Automation

    Agent Mail ships: the inbox becomes a front door for the agent

    Email a task from your sign-in address and the agent replies with what it would run and where. Nothing starts until you confirm, and writes still queue behind the normal approval gate.

  52. Platform

    DV360 brought to Google Ads parity — targeting, deals, audiences and reach forecasting

    Display & Video 360 coverage widened from reads to the full working surface: insertion orders, line items, targeting, deals, first-party audiences and native reach forecasting.

  53. Platform

    Campaign Manager 360 joins natively, with writes fenced to five trafficking collections

    CM360 joins natively: the full reporting lifecycle, campaign, placement, creative and Floodlight audits, plus a deliberately narrow trafficking write surface.

  54. Platform

    Shopify — three connection tiers, from domain-only to approval-gated store writes

    Connect a store with just its domain for catalog access, add an Admin token for orders and customers, or grant write scopes for approval-gated product and inventory changes.

  55. Models

    Bring your own model — route agent LLM calls to your own endpoint

    Point every agent at an OpenAI-compatible endpoint you control. LLM tokens on it are zero-rated in credits; a platform-infrastructure fee still covers the compute of running each task.

  56. Platform

    Meta Ads and TikTok Ads — native, cross-channel in one run

    Both platforms connect natively through a compact gateway-tool design that reaches the whole API surface by parameter, rather than a fixed list of narrow endpoints.

  57. Platform

    AdSense, AdMob and Google Ad Manager — the publisher side

    Three monetization platforms connect read-only: site earnings and RPM, app revenue and mediation waterfalls, and publisher ad-server yield.

  58. Integrations

    Composio — 250+ apps reachable, with writes behind the approval gate

    Connect Slack, Gmail, HubSpot, Notion and hundreds more through one API key, with external-connector writes gated by the same approval step as ad-platform changes.

  59. Platform

    Search Console — organic and paid answerable in one question

    Query performance, impressions, CTR and position connect read-only, so organic rankings can be cross-referenced against Google Ads keyword spend.