Buddy Levels Up: Claude Fable 5, Runtime Tag X-Ray Vision, and One Session Across Every Tool

Buddy started as a Google Ads agent that could read an account and tell you the truth about it. Over the last several weeks he has become something more: an agent with a frontier-class brain, a real browser he can drive, a memory that compounds across sessions, and a tool suite that finally behaves like one product. This post is the changelog — everything that shipped, what it does, and why it matters if you spend money on ads.

A frontier brain: Claude Fable 5 is in the picker

The headline: Claude Fable 5 — Anthropic's newest frontier model — is now available to every Buddy user, right in the model picker next to Opus 4.8, Sonnet, and Haiku. No waitlist, no special tier.

Fable 5 belongs to the same frontier generation as Anthropic's Mythos-class models, and it thinks differently than anything we have run before. Instead of a fixed "thinking budget" you toggle on and off, Fable 5 uses always-on adaptive reasoning — it decides how hard to think based on the question. We wired Buddy's backend to steer that with an effort dial: simple reads run light and fast, full audits and strategy work run at maximum depth. You never configure anything; Buddy picks the right depth per request.

Under the hood this took real work. The newest frontier models reject old-style sampling parameters entirely, and we found (and fixed) a bug where bring-your-own-key calls were sending a hardcoded temperature that these models refuse. That fix shipped across the entire tool suite, so Fable 5 and Opus 4.8 now work everywhere — in Buddy and in every satellite tool. The backend is Mythos-ready too: as frontier availability expands, Buddy's model layer already knows how to drive it.

X-ray vision: Buddy now verifies what your site actually fires

This is my favorite shipment of the batch. Ask Buddy to "run a tag audit on example.com" and he no longer just reads the page HTML and guesses. He spins up a real headless Chrome browser at Cloudflare's edge, loads your homepage and your highest-value pages (contact, pricing, checkout), and watches every single network request the site fires.

Why that matters: most modern sites run their tags through Google Tag Manager. A static HTML scan can see the GTM container, but it cannot see what's configured inside it — the GA4 tag, the Google Ads conversion tag, the remarketing pixel. Those only exist at runtime. Buddy now sees all of it:

• What actually fired on the wire — live GA4 collect hits with their measurement IDs, Google Ads conversion and remarketing loads, Meta, Microsoft, TikTok, LinkedIn, Pinterest, and eight more platforms.
• The live data layer — the real events your site pushed on load, not the ones the source code promises.
• The failure case nothing else catches — a tag that is present in the HTML but never fires. Broken, blocked, or consent-gated into silence. In testing we found exactly this on a real site: GTM in the page source, zero tag requests at runtime. Money was being spent against tracking that did not exist.

Every audit returns a 0–100 tracking health score, per-platform coverage across the pages scanned, duplicate-ID detection (two GA4 IDs means double-counted conversions), Consent Mode and CMP posture, and a ranked issue list — delivered in plain language with the what, the why, and the so-what.

And because the browser runs server-side at the edge, the experience is identical on the iOS and Android apps. Your phone sends one chat message; Cloudflare does the heavy lifting; you get the findings. Zero battery, zero data, zero waiting on your device.

Prefer a dedicated workbench? The standalone Marketing Analytics Auditor runs the full 100+ point version with HAR imports, GA4 property audits, GTM container inspection, consent simulation, and funnel checks.

A memory that compounds

Buddy has always remembered conversations. Now he learns from them. Several brain systems went from "built" to "fully wired" in this release:

• Automatic memory capture. Durable facts about you and your business — your market, your margins, the campaign you renamed last week — are extracted and stored as you talk, then recalled semantically when they matter.
• A closed learning loop. When Buddy recommends a budget change, he now checks back on the outcome and carries what worked (and what did not) into future recommendations.
• Competitor persistence. When research turns up your competitors, they are remembered — next month's audit compares against them automatically.
• Self-reflection. High-stakes responses — audits, strategy, anything touching money — get an internal critique pass before you see them. Claims are checked against the soul rubric: no fabricated numbers, no invented receipts, every figure sourced from a real tool result.

One session, every tool, every device

The AHMEEGO suite is eleven tools — Buddy, the Google Ads Builder, Keyword Analyzer, Analytics Auditor, 250-Point Audit Engine, Social Media Builder, Writing Agent, and more. They used to behave like eleven separate products: a key entered in one tool died when the tab closed, and your model preference reset everywhere, constantly. On the mobile apps, where every tool is its own page, this was a daily paper cut.

Now there is one shared session. Sign in once, enter an API key once, pick your model once — and it follows you across every tool, every visit, and both app stores. Each tool stays focused and independent (Buddy is a Google Ads operator, not a kitchen sink), but the plumbing underneath is finally unified.

A hundred small things that add up

The release also carries a long tail of fixes and polish, the kind users feel before they can name:

• Automatic light and dark mode. Buddy follows your device theme by default, with a manual override. The mobile browser chrome tints to match.
• Confirmations that work in any language. Buddy operates accounts in English, German, Czech, and beyond. The confirm flow now understands "ano", "ja", and "jdeme na to" as readily as "yes" — the multi-turn CONFIRM loop that plagued non-English sessions is gone.
• Currency-safe money handling. Budget changes are sanity-guarded and currency-aware, whether the account runs in dollars, euros, or koruna.
• Read-after-write verification. When Buddy says a budget changed, it is because he read it back from the API after writing it — not because the request returned 200.
• Real landing-page screenshots. The same edge browser now powers visual landing-page reviews — Buddy screenshots your page post-hydration and runs a conversion-focused UX critique against the image.

Watch it in action

Try it

Everything in this post is live right now. Open Buddy, pick Fable 5 from the model dropdown, connect your Google Ads account — or just paste a domain and say "run a tag audit."

What's next

The roadmap from here: long-horizon jobs (kick off a deep audit, close the app, get notified when it is done), automated quality evals against golden transcripts, and richer runtime simulation — consent-state testing and funnel walks inside Buddy himself. If there is something you want Buddy to do, tell me. The best features in this release started as user complaints.

— John