01. Analysis
The way most AI marketing tool roundups are written, you're supposed to read to the end and pick one. That's not how stacks work.
The five open-source AI marketing projects that keep surfacing in our GitHub sweeps aren't competing for the same job. They sit at different layers — analytics and intelligence at the bottom, strategic frameworks above that, content execution in the middle, video production at the top. Treating them as alternatives is like comparing a database to a CMS and asking which one you need. You probably need both, in the right order.
This roundup maps those layers explicitly, reports on which tools are actually gaining momentum right now, and tells you which configuration makes sense depending on where your marketing operation stands today.
How we researched this
On June 15, 2026, we re-ran ToolSift's automated research pipeline against the ai-for-marketing topic. The sweep returned the same five GitHub repositories that appeared in our June 8 analysis — with updated star counts and commit timestamps. Reddit returned zero posts above our minimum score threshold. Hacker News returned zero stories above our 15-point floor. Official pricing pages for Jasper AI, Copy.ai, Surfer SEO, and HubSpot AI again returned no content.
For the third consecutive research run, GitHub is the only live signal available. I've started to read that as information: the AI marketing community is building and shipping on GitHub, not debating on Reddit or making the HN front page. The absence of community noise doesn't mean the tools aren't being used — it means the users are practitioners, not commentators.
The seven-day gap between our June 8 and June 15 sweeps also gives us something genuinely new: week-over-week momentum data. Star counts on their own tell you historical interest. Growth rates tell you which tools are in active adoption right now. I'll lean on that data throughout.
The stack, layer by layer
Think of AI marketing infrastructure as four layers: intelligence (knowing what your market wants and what's working), strategy (deciding what to do with that knowledge), content execution (producing the output), and video production (formatting that output for short-form channels). The five tools in our research set map cleanly onto these layers — and that mapping is the insight that most single-tool roundups miss.
Layer 1 — Intelligence: ikatsov/tensor-house — 1,446 stars
The foundation of any serious marketing operation is knowing things: which customer segments matter, which prices the market will bear, which campaigns are actually driving revenue. Tensor-house is the only repository in our data set that addresses this layer directly.
At 1,446 stars (up 5 from 1,441 on June 8, a +0.35% weekly growth rate), tensor-house is a collection of reference Jupyter notebooks built around enterprise ML use cases. The marketing-relevant notebooks cover multi-touch attribution modeling, customer lifetime value estimation, demand forecasting, and promotional pricing optimization. These are problems that growth-stage marketing teams typically solve by either hiring a data science consultant or buying an analytics platform they then underuse.
The 0.35% weekly growth rate is the lowest in this dataset. I read that as a signal of maturity, not stagnation. Tensor-house is a reference implementation library, and those accrete stars slowly from practitioners who find the notebooks while solving specific technical problems. It isn't the kind of tool a marketing influencer tweets about. The 1,446-star base represents years of steady discovery by people actually using ML in marketing analytics.
What this layer gives the stack: Ground truth. If you know your customer LTV distribution, your attribution model, and which price points elastically respond to promotions, every downstream decision becomes more precise. Without this layer, you're producing content and videos based on intuition and optimism.
What it requires: At least one person who can read Python and Jupyter notebooks. Your own data — enough transaction and campaign history to train the models. Some patience with notebook environments that aren't production-ready out of the box.
Last updated: June 12, 2026.
Layer 2 — Strategy: BrianRWagner/ai-marketing-claude-code-skills — 324 stars
Above the analytics layer sits strategic frameworks: positioning documents, messaging architectures, campaign planning structures. BrianRWagner's repository provides structured AI execution frameworks for exactly this layer — and it's growing at 1.89% weekly (up 6 from 318, second-fastest in the dataset).
The repository is described as "marketing frameworks that AI actually executes" and is explicitly "designed for Claude Code." That design choice matters. These aren't generic prompts you paste into any chat interface. They exploit Claude Code's tool-use and project context architecture to walk an AI agent through a structured framework: audience definition, competitive positioning, message hierarchy, proof point selection. The output is not a piece of content — it's a strategic document that should inform everything the content layer produces.
The 1.89% weekly growth rate is notable for a sub-350-star repository in a narrow niche. Small technical repos in specific tooling ecosystems can plateau quickly once the early adopter community has found them. That BrianRWagner continues to grow at this rate suggests it's still being discovered by new Claude Code users, not just circulating among the same early-adopter community.
What this layer gives the stack: The positioning foundation that prevents AI-generated content from being generic. If you're running Affitor or any other content workflow without a positioning framework underneath it, your output will sound like every other AI-produced content in your space. BrianRWagner's frameworks force the AI to work through your specific market position before it writes anything.
What it requires: Claude Code. The skills are designed for that specific environment and don't fully transfer to other AI assistants. If ChatGPT or Gemini is your primary tool, this layer's value is significantly reduced.
Last updated: June 14, 2026.
Layer 3 — Content execution: Affitor/affiliate-skills — 463 stars
Affitor is the fastest-growing repository in our June 15 dataset by a significant margin. At 463 stars (up 22 from 441 on June 8), it grew 4.99% in a single week — the only tool in this data set showing that rate of acceleration. For a niche open-source repository serving affiliate marketers, 22 new stars in 7 days is a signal worth taking seriously.
The repository provides 50 structured agent skills that cover the affiliate content production cycle end to end: trend research, data-backed post writing, infographic generation, landing page creation, deployment. It explicitly supports Claude Code, ChatGPT, Gemini, Cursor, Windsurf, and others — positioning it as model-agnostic at the expense of the deep integration that BrianRWagner achieves specifically within Claude Code.
In the stack framing, Affitor lives at the content execution layer: it takes a strategic direction (ideally informed by tensor-house data and a BrianRWagner positioning framework) and systematizes the production of that content at scale. The 50-skill count is ambitious. I expect most practitioners are using a subset of 10–15 validated skills rather than all 50 — but the breadth gives you room to customize by niche and format.
The acceleration from 441 to 463 stars in one week deserves attention. Affitor's prior growth was more gradual based on our May data. Something broke the tool into a new audience between June 8 and June 15 — possibly a newsletter mention, a social post from a practitioner, or an update that resolved a usability barrier. Reddit and HN returned no relevant threads, so I can't confirm which. But the momentum is real and the next week's data will tell us if it holds.
What this layer gives the stack: Volume content execution. If your strategy is to produce a large body of content — affiliate review posts, programmatic landing pages, niche comparison articles — Affitor provides the repeatable agent workflow that makes that possible without manual prompt construction for every piece.
What it requires: An AI coding assistant (Claude Code for best results, though the model-agnostic design gives you options). Budget for ongoing API costs — the per-piece cost is real and variable depending on content depth and which skills you invoke.
Last updated: June 15, 2026.
Layer 4 — Video production: YILS-LIN/short-video-factory — 4,084 stars
The most-starred AI marketing repository in our sweep by a wide margin, and the one most likely to be underused by English-speaking teams because of its Chinese-first documentation. Short-video-factory added 35 stars between June 8 and June 15 (+0.86% weekly growth), continuing steady accumulation from a large base.
The tool generates product marketing short videos in batch — desktop-native, cross-platform, designed for volume. The repository description translates to "one-click generation of product marketing and general content short videos, AI batch automatic editing." The emphasis on "batch" is meaningful: this is not a tool for producing one polished video per session. It is designed for teams who need to turn product assets into 20, 50, or 200 short-form clips without manual editing per clip.
At the top of the marketing stack sits production: taking the content that strategy and execution have defined, and turning it into format-specific deliverables. Short-video-factory owns that layer for video — the highest-demand short-form format in most consumer marketing contexts right now.
The 0.86% weekly growth rate puts short-video-factory in a mature, stable growth pattern. At 4,084 stars, it has by far the most evidence of real-world use among this cohort. The Chinese-first documentation means its traction comes disproportionately from practitioners, not from English-language blog coverage.
What this layer gives the stack: The video production capability that subscription SaaS tools charge $200–$400/month for. If your content execution layer (Affitor) is producing scripts and asset briefs, short-video-factory can batch-process those into video output without a per-video SaaS fee.
What it requires: A desktop installation (Windows or Mac). Willingness to work through Chinese documentation or invest time translating it. Validation that your specific video format and asset types work in the batch workflow before relying on it at scale.
Last updated: June 15, 2026.
Honorable mention: re50urces/Awesome-AI — 850 stars
Awesome-AI is a curated list of AI tools, not an executable component of the stack. At 850 stars and updated June 15, it has real discovery value: if you've assembled the four core layers above and want to find specialized additions — SEO tooling, social scheduling integrations, analytics connectors — this is the repository to browse.
I'm including it because it keeps surfacing in our research sweeps and some readers will arrive here wondering whether they've missed a better tool. The answer is no — Awesome-AI is a starting point for exploration, not a substitute for any of the functional layers above.
Week-over-week momentum table
| Tool | June 8 stars | June 15 stars | 7-day delta | Weekly growth |
|---|---|---|---|---|
| short-video-factory | 4,049 | 4,084 | +35 | +0.86% |
| tensor-house | 1,441 | 1,446 | +5 | +0.35% |
| Awesome-AI | — | 850 | — | — |
| Affitor/affiliate-skills | 441 | 463 | +22 | +4.99% |
| BrianRWagner skills | 318 | 324 | +6 | +1.89% |
The headline number is Affitor's 4.99% weekly growth. In any given week, a sub-500-star repository in a niche space growing by 22 stars is notable. Whether this represents a one-time spike or genuine sustained acceleration is the question I'm watching. For context: a tool growing at 5% per week would more than double in five weeks. That doesn't happen by accident.
What I'd use and why
The full stack for a mid-size e-commerce operation: tensor-house for attribution and customer segmentation, BrianRWagner for positioning frameworks, Affitor for content execution, short-video-factory for product video. This is the most complete configuration and the most expensive in time and technical overhead — you need a data analyst, a Claude Code user, and someone willing to install and maintain a desktop application. The upside is a marketing intelligence and production system with no recurring SaaS cost.
The lean configuration for a solo affiliate operator: Affitor plus short-video-factory, skipping the analytics and strategy layers. Affitor handles content research through deployment; short-video-factory handles the video component. The risk is producing content without a validated positioning foundation — you're betting on volume to find what resonates rather than analyzing your way there first. At Affitor's current growth rate, this configuration is also the one most likely to have fresh documentation and active maintainer support.
The strategy-first configuration for a marketing team without technical resources: BrianRWagner's skill pack inside Claude Code, used to structure positioning and campaign planning before any content is created. No installation, no data infrastructure, no batch workflows. Just better AI-assisted strategic thinking applied to deliverables your team already produces. This is the lowest-friction entry point if Claude Code is already your working environment.
For a data-heavy growth team that has outgrown spreadsheet attribution: tensor-house as a standalone investment, specifically the multi-touch attribution and LTV modeling notebooks. These are the two capabilities that most growth teams with 12+ months of data need and rarely have in-house. Tensor-house gives you a credible starting point without buying an enterprise analytics platform.
Limitations
No community signal, again. Reddit and HN returned zero relevant posts for the third consecutive run. I cannot tell you what practitioners are complaining about, which bugs are blocking adoption, or which use cases are working better than the documentation suggests. All analysis is based on GitHub metadata, repository descriptions, and week-over-week star deltas.
One week of momentum data is indicative, not conclusive. Affitor's 4.99% growth could be a one-time spike from a single mention, or it could represent genuine acceleration. I won't know which until the June 22 sweep. I'm citing the number because it's the most notable signal in this data set, not because I can explain its cause.
Stack integration isn't documented anywhere. The four-layer framing is my synthesis, not the maintainers'. There are no official integrations between tensor-house, BrianRWagner, Affitor, and short-video-factory. The data handoff between layers — from tensor-house attribution data to a BrianRWagner positioning framework, for example — requires custom work. I'm describing compatible design philosophies and complementary jobs-to-be-done, not plug-and-play connections.
Chinese documentation is a real barrier for short-video-factory. I have not personally run the tool. English-speaking teams adopting it should budget significant setup time and not treat the documentation gap as a minor issue.
Bottom line
The insight this week isn't new tools — it's a new frame. These five repositories aren't alternatives; they're layers. And the layer experiencing the most momentum right now is Affitor, up nearly 5% in a single week in a way that suggests active new audience discovery, not just steady organic accrual.
If you've been treating AI marketing tools as a "pick one" decision, the stack frame is the corrective. The open-source coverage now spans intelligence, strategy, content execution, and video production. The question is which layers you actually need and how much technical overhead you can absorb to get them running. The analytics and strategy layers (tensor-house, BrianRWagner) require the most setup. The execution layers (Affitor, short-video-factory) are more accessible but carry ongoing API and maintenance costs.
Building all four layers gets you a serious marketing capability with no recurring SaaS subscription. Building one or two gets you started while you evaluate whether the rest is worth the overhead. Neither answer is wrong. The wrong answer is treating this as a competition between five tools when it's actually a question about which layers of one stack you're ready to operate.