Skip to content
Tegra/Agents/Changelog

The agents. The record.

Framework upgrades, new tools, and quality improvements — dated and linked to the agent they belong to. Explore what shipped, how the work changed, and the detail behind each release.

The release record. Every change has a date.

Search the latest 15 published entries. Filters update the charts, agent summary, and release notes together.

PUBLISHED ENTRIES
15Agents-only release feed
LOADED HERE
15Most recent entries available to filter
MATCHING FILTERS
15The sample used by the charts below
LATEST MATCH
Oct 20, 20261.3.0

15 matching releases · 15 loaded · 15 published

Recent changes in this view

The latest matching published release notes. Open a record for its full entry below.

Latest 3 of 15 matching loaded published releases · 15 loaded of 15 published in the full agent feed.

  1. feature
  2. feature
  3. improvement

Source: Published agents-only release feed · Publication records with their original source dates and release identities.

Release activity · current agents

Only current public agents, across up to six represented months, oldest to newest. Counts describe published changes, not agent performance. Historical agent releases remain in the records below.

Google Ads AgentMeta Ads Agent

Scroll sideways to see every period.

Jun 26Google Ads Agent: 1 releases; Meta Ads Agent: 0 releases
Jul 26Google Ads Agent: 0 releases; Meta Ads Agent: 2 releases
Oct 26Google Ads Agent: 2 releases; Meta Ads Agent: 2 releases

0–2 · current-agent published entries in this sample. All periods share this scale.

Read the values
Release activity · current agents · source values
PeriodGoogle Ads AgentMeta Ads Agent
Jun 261 releases0 releases
Jul 260 releases2 releases
Oct 262 releases2 releases

Source: Published agents-only release feed · 7 plotted entries for current public agents out of 15 matching loaded records. Up to six represented months are shown. Historical agent records remain in the release notes below; this is not the full historical cadence.

What changed

Categories retain their published labels. A fix is a shipped correction, not an unresolved incident.

Launch7 releases
Feature5 releases
Improvement3 releases

0–7 · matching releases per category

Source: Published agents-only release feed · Charts use 15 matching records from the latest 15 loaded of 15 published entries; this is not the full historical cadence.

Agent release summary · matching loaded entries
AgentEntriesLatest in this sample
Google Ads Agent
1.4.0 · Oct 20, 2026
Meta Ads Agent
1.3.0 · Oct 20, 2026

Source: Published agents-only release feed · Entry bars share a 0–4 scale. Older or retired agent keys can remain in the dated release record below.

featuremajor1.3.0Meta Ads Agent

Meta Ads Agent v1.3.0: the production system's 3.0 creative doctrine

The Meta Ads Agent now runs the creative and media-buying doctrine of the Meta Ads AI Agentic System 3.0: multi-text feed copy for every static, a deeper competitor spy, four-part angles, and test plans written in multiples of your target CPA.

Read 7 release highlights
  • **5 x 5 x 5 feed copy.** Static briefs and the new Feed Copy Pack preset ship 5 primary texts, 5 headlines, and 5 descriptions per image, each tagged by format, as a downloadable CSV next to the brief pack.
  • **Competitor spy that separates proven from killed.** The Ad Library search now reads active and inactive ads, buckets each by how long the brand kept it running, counts copies, decodes each template, and writes clone briefs only from templates kept for months. — Searches run in parallel inside a time budget, so the run finishes even when the library is slow.
  • **Four-part angles.** Every angle carries a Big Idea, Unique Mechanism, Key Benefits, and Objection Handling, built from personas and desires in your customers' words and filtered for Visualizable, Falsifiable, and Unique.
  • **Launch plans in multiples of your target CPA:** cells your budget supports, daily budgets and bid caps, and the kill, hold, and scale rules the production system uses.
  • **Funnel diagnosis through the bid-cap lens,** with trajectory shapes and a named bottleneck behind every verdict.
  • **Video scripts** with ten openers written after the script, typed with canonical hook types, and a SHIP / FIX / KILL roast.
  • **Templates refreshed,** including the new Feed Copy Pack and the Launch Plan + Test Matrix with a required target CPA.
featuremajor1.4.0Google Ads Agent

Google Ads Agent v1.4.0: builds that finish, on the production system's current rules

The Google Ads Agent now builds Search, Performance Max, and Demand Gen files in one validated step, so research-backed builds deliver their files inside a single run. Every build follows the production system's current launch rules, and every file arrives Paused.

Read 8 release highlights
  • **Builds deliver their files.** Search, PMax, and Demand Gen builds hand the agent's plan to a builder that writes the exact Google Ads Editor files, checks them, and returns them in one step. — A full research-backed Search build that used to run out of time now completes with its files.
  • **No target, no launch.** Non-brand campaigns launch on a target-based bid strategy calculated from your AOV and margin (break-even ROAS = 1 / margin), brand defense runs separately on Manual CPC, and Maximize Clicks is no longer recommended. — Without your numbers, the files are labeled NOT READY TO LAUNCH instead of guessing.
  • **Keywords chosen by what you can afford per click.** Keyword selection uses CPC headroom against your economics instead of fixed volume and CPC floors.
  • **RSAs where each ad wins a different click,** checked for cross-ad-group overlap, repeated lead words, restated descriptions, and banned characters, with accents allowed in non-English markets.
  • **Full PMax asset groups:** 15 headlines, 5 long headlines, 5 descriptions, 20-50 search themes, image briefs that go deep on one angle, video briefs, and the launch settings checklist.
  • **New Demand Gen Kit mode and preset:** one ad group per persona and angle, image ratios Demand Gen accepts, tCPA from your target, the kill rule, and step-by-step replacement of a disapproved ad.
  • **Post-launch cuts use a cut floor tied to your target** (max($30, 3 x target CPA)); terms below it are held, not cut.
  • **Shopping feed rules updated:** anchor keyword in the first 35 characters, standalone short titles, real colours only, margin and velocity custom labels.
improvementminor1.2.0Meta Ads Agent

Meta Ads Agent v1.2.0: GPT-6.1 Sol, angles that differ from each other, and a brief pack that stands on its own

The Meta Ads Agent now runs on GPT-6.1 Sol, the newest flagship model. Its three angles have to be three different ideas, and the downloadable brief pack carries everything production needs without the chat next to it.

Read 6 release highlights
  • GPT-6.1 Sol is the default model and the Sol option in the model picker, and runs on it cost less: credits are now charged at GPT-6.1 Sol's published rates, which are lower than the previous Sol's.
  • The picker now offers two tiers: Sol, and Luna for fast drafts at about a tenth of Sol's price. — Terra has been retired from the picker, since Sol is now both the newer model and the cheaper one.
  • Surviving angles must differ from each other, not only from competitors. — Each one owns its own customer tension, proposition, and visual execution; attaching the same product facts to three angles no longer passes the Unique check. Hooks follow the same rule.
  • The creative brief pack download now includes the exact on-image copy and aspect ratio for every brief, so a designer can produce from the file alone.
  • The agent states plainly which files were delivered instead of describing a download as unavailable.
  • A step that fails mid-run no longer ends the run. — The agent corrects the step and carries on.
improvementminor1.3.0Google Ads Agent

Google Ads Agent v1.3.0: GPT-6.1 Sol, and builds that stay inside the scope you set

The Google Ads Agent now runs on GPT-6.1 Sol, the newest flagship model, and it holds to the scope you give it. Tell it to skip external research or to build exactly two RSAs per ad group and that is what you get, with files that match what it validated.

Read 6 release highlights
  • GPT-6.1 Sol is the default model and the Sol option in the model picker, and runs on it cost less: credits are now charged at GPT-6.1 Sol's published rates, which are lower than the previous Sol's.
  • The picker now offers two tiers: Sol, and Luna for fast drafts at about a tenth of Sol's price. — Terra has been retired from the picker, since Sol is now both the newer model and the cheaper one.
  • Your stated scope now overrides the build defaults. — "Skip keyword research", "use the context I pasted", and fixed counts for campaigns, ad groups, keywords, or RSAs are followed exactly instead of being expanded to the full build protocol.
  • Build runs produce the files first and keep the closing write-up short, so more of each run goes to deliverables you can import.
  • CSV downloads are byte-identical to the content the agent validated, including the final line ending, so the Import QA summary describes the exact file you download.
  • A step that fails mid-run no longer ends the run. — The agent corrects the step and carries on, and the tool card says so instead of reporting that the whole run failed.
featuremajor1.1.0Meta Ads Agent

Meta Ads Agent v1.1.0 - Full-System Replication

The agent now replicates the entire Meta Ads AI Agentic System surface (minus rendering and account access): nine modes covering research, angles, the Big-Idea UMP/UMS pipeline, static brief batches, video scenarios with an adversarial roast, live Meta Ad Library competitor spy, the 7-step funnel diagnosis, launch planning, and copy audits.

Read 4 release highlights
  • 9 modes — Research to launch plan - the full system surface, hosted
  • Video scenarios, roasted — Spine, beats, persona-locked VO, storyboard - adversarial panel verdict included
  • Live Ad Library spy — Coverage maps, white space, and faithful template rip-off briefs
  • 7-step diagnosis — AD vs FUNNEL vs OFFER vs TARGETS, with max 3 routed actions
launchmajor1.0.0Meta Ads Agent

Meta Ads Agent v1.0.0 - Launch

New public agent: the Meta Ads AI Agentic System's creative doctrine in one hosted run - psychological personas, angles through the 3-Rule Filter (kill list included), a production-shaped static brief, and Found-Not-Written video hooks from your product URL and reviews.

Read 3 release highlights
  • 4th public agent — Personas, filtered angles, static briefs, and hooks from a product URL
  • The kill list — See which angle candidates failed the 3-Rule Filter and why
  • [NEEDS SOURCE] honesty — The agent never invents product claims
improvementminor2.5.0

Creative Brief Builder v2.5.0: verbatim-first hooks and headlines

The Creative Brief Builder now runs a Found-Not-Written pass before authoring the copy slots where marketer-speak kills performance fastest: image headlines, copy overlays, video hooks, thumbnail text, and UGC script hooks. Real customer language gets tried first; fresh copy is the fallback, not the default.

Read 3 release highlights
  • Verbatim-first authoring: the agent collects real customer phrases from the reviews, scraped product pages, and audience context you provide, ranks them by concreteness and source quality, and tries the strongest verbatim in each hook and headline slot before writing anything original.
  • Honest sourcing labels: copy lifted from a real customer phrase is marked [VERIFIED] with its source; briefs produced without any customer language are labeled so you know exactly what pasting 5-10 real reviews would improve on the next run.
  • Fresh copy that is written rather than found must still clear the 3-Rule Filter and now also has to echo the texture of the extracted customer language instead of defaulting to marketing diction.
featureminor1.2.0

Landing Page Builder v1.2.0: five hard layout gates and verbatim-first copy

Every page the Landing Page Builder ships now passes five hard layout gates before it is marked complete - each one invented after a real page shipped broken in production. And before writing any headline, the agent now runs a verbatim-first pass: it tries real customer language from your reviews before writing a single line of its own.

Read 3 release highlights
  • Five pre-deploy layout gates run on every rendered page: no competitor commerce links, no affiliate-template disclosure language, canonical packshot in the final CTA block, no dead CTA buttons, and no build-pipeline metadata visible to readers. — A five-line gate summary prints above every build completion so you can verify before deploying.
  • Three of the gates now also run as deterministic scanners inside the expert review panel - dead buttons, disclosure leaks, and technical metadata leaks are caught mechanically on the rendered HTML, not just by model judgment.
  • Found-Not-Written copy pass: the agent extracts verbatim phrases from the reviews and customer language you provide, tries the strongest verbatim for each headline slot first, and only writes fresh copy when no real phrase fits. — Copy written without any customer language is labeled honestly so you know what one pasted review batch would improve.
featureminor1.2.0Google Ads Agent

Google Ads Agent v1.2.0: post-launch optimization with directional discipline

The Google Ads Agent now optimizes live campaigns, not just new builds. The new SEARCH OPTIMIZE mode reads your search term report and walks three directional passes - sense, add, remove - borrowed straight from the production system that manages live spend across 21 brands. Adds and removals are never mixed in one list, because mixed lists are how mass-deletion mistakes happen.

Read 3 release highlights
  • New SEARCH OPTIMIZE mode: bring a live campaign and a search term report, and the agent runs three directional passes. — MINE surfaces winners, losers, and drift as a findings table with evidence per row. GROW delivers add-only Editor-ready CSVs: new exact-match keywords from proven search-term winners and new negatives from proven losers. HARDEN delivers a separate remove-only pruning plan with a reason code and the evidence for every line.
  • Hard evidence bars on every recommendation: every metric names its window (30, 90, or 180 days), nothing is removed on Google's "Low" labels alone, and anything with a conversion in the lookback window never appears on a removal list.
  • Cadence guidance ships with every run: MINE + GROW as a weekly loop, HARDEN as a quarterly cleanup, with the next step named in the closing actions.
launchmajor1.0.0

Creative Brief Builder - Image Briefs, Video Scripts, and 4-Layer Architecture

Builds scored Demand Gen image briefs, YouTube video scripts across 5 formats, and a complete 4-Layer Campaign Architecture from a shared emotional truth foundation.

Read 4 release highlights
  • New Demand Gen + YouTube creative agent
  • 100-point headline scoring rubric
  • 5 YouTube script formats
  • 4-Layer Campaign Architecture for scaling
launchmajor1.0.0

Competitor Researcher - Blue Ocean Scoring and Creative Briefs

Reverse-engineers competitors, scores differentiation on a Blue Ocean 0-100 scale, declares the creative pole (Organic vs DR), and outputs scored creative briefs with UMP/UMS mechanism specificity.

Read 4 release highlights
  • New competitive research + creative strategy agent
  • Blue Ocean differentiation scoring (0-100)
  • UMP/UMS mechanism specificity in briefs
  • Organic vs DR creative pole calibration
launchmajor1.0.0

Landing Page Builder - Product URL to Presell Pages in One Pass

Give the agent a product URL and it scrapes, researches keywords, discovers and scores presell angles, writes copy with 15+ frameworks, and builds up to 4 downloadable HTML pages.

Read 4 release highlights
  • New URL-to-presell-pages agent
  • 15+ copy frameworks (Schwartz, Cialdini, Masterson)
  • 3-phase pipeline: DISCOVER → WRITE → BUILD
  • Downloadable HTML with responsive CSS and Schema.org
launchmajor1.0.0

Shopping Feed Optimizer - Full Feed Rewrite Pipeline

Upload a Merchant Center or Shopify feed and get keyword research, BOFU/MOFU/TOFU segmentation, multi-version rewrites with unique IDs, and a 32-column canonical CSV ready to upload.

Read 4 release highlights
  • New full-pipeline feed optimization agent
  • 5-phase workflow: Analysis → Research → Segmentation → Rewriting → Delivery
  • 32-column canonical CSV output
  • DataForSEO keyword research built in
launchmajor1.0.0

PMax Campaign Builder - Funnel-Aware PMax & Shopping Audit Agent

Deep diagnostic of PMax and Shopping setups across funnel layers, feed quality, asset group signals, segment density, and competitive gaps - returns a ranked optimization backlog.

Read 4 release highlights
  • New PMax + Shopping funnel-aware audit agent
  • 152-point feed quality standard
  • Ranked optimization backlog by likely impact
  • BOFU/MOFU/TOFU funnel layer scoring
launchmajor1.0.0

Search Campaign Builder - Structure, Negatives, Extensions, RSAs

Turns offer context into a launch-ready Search campaign: 5-tier negative keywords, scored RSAs across 8 hook families, ad extensions, and a funnel-aware match strategy.

Read 4 release highlights
  • New Search structure + RSA generation agent
  • 5-tier negative keyword framework
  • 8 hook families across 5 headline types
  • Upload-ready launch sheets

Showing the latest 15 of 15 published agent entries. Dates use UTC.

Subscribe via RSS
DISCUSS YOUR CHANNEL AND IMPLEMENTATION

Founder-led execution on your accounts.
Or run them yourself.

Two ways to work with Tegra. Book a 30-minute discovery call with a founder to see whether your account is a fit for full-funnel done-for-you execution. Or skip the call entirely — buy a productized system and self-deploy in a day.

Book a discovery call→Browse products · $19,997self-deploy →
✓ No credit card✓ 30 min on a call✓ Reply in 4h