What’s new
Every change that reached production, newest first — what it changes for you, not what it changed in the code. Releases are dated: the agent ships continuously, so the day is the version.
Tag Manager releases go through
A Tag Manager release across many containers no longer stops at its first step on a container that is ready for it.
Changes prepared in the background
Ask for one change across hundreds of containers, properties, sites, repositories or ad accounts, Google Ads, Meta, TikTok, Snapchat and LinkedIn included, and review it all on a single card; anything that would start spending waits for your tick. Writes left with an unknown outcome can now be closed from Approvals & runs once you have looked yourself.
Confirmations that match the change
You confirm only what the agent is about to do, named for what it is, and only a deletion depends on your deletions setting. One approval now carries a change across many containers step by step, each stopping alone where something fails, and a long task ends each reply with where it stands.
Tag Manager additions that only add
Adding a setting to a Tag Manager tag or variable now keeps everything already there, whether you ask in the chat or from your own AI assistant, and whatever changed before you approved. The review card shows each table's rows before and after, and warns when one would lose some.
A console to start from
Signing in now opens a console home: what waits for your approval, your connections and recent activity at a glance, and every part of YAAPI one click away. A first visit walks you through setup step by step, whether you work from your own AI assistant, from the chat here, or both.
The work done, on record
YAAPI now keeps a lasting count of the changes it makes on your accounts, per organisation and without naming anyone, so the work done for you can be measured well beyond the log windows. The privacy page says exactly what that record holds.
LinkedIn Ads, ready for campaigns
LinkedIn Ads now appears among the platforms you can connect, and the agent can sponsor a post your page already published, read audiences by job title or company size, and pause or resume a campaign on your word. A plain yes confirms a change, and a fix that would widen who sees your ads asks you again.
Central changes, local settings kept
A change pushed from a reference Tag Manager container now lands on top of each container’s own settings instead of erasing them: what a market added locally stays, and the review card says per item what stays, what is added and what changes. When the reference and a container disagree on a value, the agent asks which should win.
Enterprise inside your company, and Google Analytics reporting in full
The Enterprise plan now runs YAAPI inside your own infrastructure, under your governance, with specialised support from integration to everyday use. In Google Analytics, funnel reports, long reports, the audience lists behind your audiences and your remaining quota are now within reach, and what the agent knows of each Google API stays complete by rule.
Names are just names, and LinkedIn joins the fleet
A rename is never mistaken for a deletion, and "Run now" on a schedule brings the same approval card a slot would. Scheduled runs come in searchable pages, archives a click away. A review card shows everything a change would write, and a finished run keeps it. LinkedIn Ads joins, prepared as on Google Ads, Meta and TikTok.
Budgets that are yours, Meta that asks once
Your Google Ads allowance is counted as Google counts it — per Cloud project, in real operations — so one team's day no longer stalls another's, and a project not yet allowed onto real accounts is told where to fix it. On Meta, an ad set or ad missing something is refused once, everything needed in one message.
Matomo Analytics joins the conversation
Connect your own Matomo — self-hosted or in Matomo's cloud — with a token from your account, and ask about its sites the way you ask about Google Analytics: visits, pages, referrers, goals and conversions over any period, plus the goals, segments, dimensions and Tag Manager containers behind the numbers. Your visitors' own data stays out of it.
Looker Studio, and who can open what
Connect Looker Studio and ask what nobody keeps a list of: every report and data source you can reach, who owns each one, and which of them are open to anyone with the link. When you want that changed, you say so and approve it like any other change.
Metabase joins the conversation
Connect your Metabase, hosted or on your own server, and ask about what lives in it the way you ask about everything else: the dashboards your teams open daily, and the answers behind them. The agent can build and mend that content, and only ever reads your data.
A schedule says what you asked for
When you say when, that is when: the hour is read in the place you named, and a plan that would run on a different rhythm than you asked for is refused rather than quietly rewritten. Each plan also says where its time zone came from, so you can spot a wrong one before approving.
Scheduled runs, and every account gets its turn
Scheduled runs arrive: ask once and the agent watches on its own, daily, weekly, monthly or on a single date you name, on any product you connected, then mails the report and lets you take it away as a spreadsheet. You approve every plan first, nothing changes without your click, and a failure says whose problem it is.
Search Console joins, and your secrets stay yours
Organic search joins the conversation, and your secrets stay yours: the agent never repeats a value it was told to keep, and the privacy page now says, item by item, what is kept about your account and for how long.
One change asks just like a thousand
A single change now waits for your approval in the conversation, exactly as a batch does — and so do a copy from one property or container to another, a mass update and a table load. The agent no longer answers a request about one product with a question about another.
Building a ChatGPT Ads campaign takes one exchange
Setting up a campaign no longer turns into a string of questions: the agent gathers what it can read, asks you once for the rest, and comes back with a complete plan. And ChatGPT Ads now appears everywhere the site says what you can connect.
ChatGPT Ads and Snapchat Ads join, and the app keeps up with you
The agent keeps its footing when a platform answers slowly or not at all, and says plainly what it could not finish.
Bulk changes as review cards, two more platforms, and the guarantees audited
More of the places your campaigns live can now be read from the same conversation, without leaving it.
Choose who answers
The chat runs on an OpenAI key as well as an Anthropic one, large approvals finish on their own, and a conversation stays in your language.
Where your work goes, on one page
The privacy page says where your work goes and on whose accounts it runs, and IT departments get a runbook of their own.
Clearer actions, on every browser
Connected assistants name the service behind each action, and the site works again on older iPhones and Macs.
A front door
YAAPI has a home page at yaapi.net, a shared computer keeps each account’s conversations apart, and the largest read-only questions run to the end.
Your word, taken literally
A change runs only when you clearly said yes, only on the account you named, and anything that erases data asks for its own permission.
Asked once, and it means something
You are asked once per decision and nothing goes ahead without your word. Links keep your place, and a long history no longer slows the app.
Nothing finishes half-done in silence
Your dbt project can be changed at scale, always by pull request; an interrupted run comes back with a one-click resume; approval requests speak your words.
Two plans, and changes at fleet scale
Reading stays free and writing joins the Pro plan. One change can cross hundreds of properties or containers as a single reviewable, reversible run.
YAAPI in your AI assistant, and approvals on a page of their own
Connect YAAPI to claude.ai, ChatGPT, Claude Code or Cursor and work from there, deciding how much it may do. What an assistant asks to delete waits on a page built to be read, and you decide item by item.
dbt Cloud, GitHub, and a deletions switch
The agent reaches your data pipeline through dbt Cloud and GitHub, and erasing anything now needs a permission you turn on per connection.
Your own Google Cloud project
Google is called through a project you own, and replies can be trimmed to the essentials.
Client accounts from the conversation
A client account can be created without leaving the conversation.
Google Ads you can rely on
Google Ads changes go through cleanly, a broken connection is reported where you can fix it, and what the app does, and under which terms, is written down.
Google Ads, and a Settings page that says what it allows
Google Ads joins the agent, and Settings shows what each connection allows before you grant it.
Five more platforms
Meta, TikTok, Snapchat, LinkedIn and HubSpot join, and nothing is changed anywhere without you.
One connection per product
Each product is connected, and revoked, on its own.
Exports that finish
Large exports report progress, resume, and land in a file.
Read-only by default, and your data on request
Read-only became the default, and everything stored about you became visible, exportable and deletable.
Nothing claimed that did not happen
Guardrails around anything irreversible, and nothing reported as done unless it actually happened.
Usable on a phone
Uploads and downloads became reliable, and the interface works on a small screen.
First production release
YAAPI Agent goes live.