Tech Stack · Resources
The AI Tool & Model Directory
The AI ecosystem looks like chaos — hundreds of products, new ones weekly, half of last year's gone. It isn't chaos. It's a stack. Every tool below fills a slot in one of the six layers we cut open in the Layered Autopsy, and the slot outlives the tool: whatever is fashionable this quarter, you will always need silicon to run on, a building to run it in, a model to call, wiring around the model, and an application on top. Learn the layers and the churn stops mattering.
This directory is organized by those layers. Each entry says what the thing is and — more usefully — when you would actually reach for it. Where a link pays us a commission, it's marked, and the disclosure above applies. The rankings and one-liners are ours; nobody bought a sentence.
---
Some links on this page are affiliate links. If you sign up or buy through one, TheCatch.AI earns a commission at no extra cost to you. We list what we would use; the commission never decides the ranking, and nothing in Bubble Watch or Economics carries an affiliate link — analysis stays clean.
17 entries
Hardware
The chips, and the places you rent them by the hour.
The chipmakers (reference — know who makes the floor you stand on)
the company whose GPUs run the overwhelming majority of AI training and inference; CUDA is the moat. Reach for it when: you're buying or renting compute and want the default that everything supports.
Chips
the credible challenger; MI300/MI325-class accelerators with large memory, running the ROCm software stack. Reach for it when: you want price-per-gigabyte-of-VRAM leverage and your stack runs on ROCm.
Chips
Google's in-house accelerator, rentable only through Google Cloud; the one major stack that isn't NVIDIA. Reach for it when: you're training on JAX or want an alternative supply curve.
Chips
Affiliate
wafer-scale chips (one giant chip instead of many small ones) sold as systems and as an inference API. Reach for it when: you want extremely fast inference on supported open models.
Chips
Affiliate
custom "LPU" inference silicon behind an API; known for very high tokens-per-second on open-weight models. Reach for it when: latency is the product and an open model is good enough.
Inference silicon
the fab that manufactures nearly every advanced AI chip; the physical chokepoint of the entire industry. Reach for it when: you're reading the supply chain, not buying anything.
Fab / reference
GPU clouds & rental (the money shelf — this is where readers actually spend)
Affiliate
AI-first GPU cloud; on-demand and reserved H100/B200-class instances plus one-click clusters. Reach for it when: you want serious training hardware without hyperscaler procurement.
GPU cloud
Affiliate
the biggest of the neoclouds; hyperscale GPU capacity, Kubernetes-native, now a public company. Reach for it when: you need real cluster scale and a negotiated commitment.
GPU cloud
Affiliate
pay-per-second GPU rental with a community cloud tier and serverless GPU endpoints. Reach for it when: you want the cheapest practical way to grab a single GPU for hours, or serverless inference on your own model.
GPU cloud
Affiliate
a marketplace of other people's GPUs, from data centers to hobbyist rigs; lowest prices, variable reliability. Reach for it when: cost matters more than SLAs — experiments, batch jobs, fine-tune runs you can restart.
GPU marketplace
Affiliate
beyond its inference API, Together rents dedicated training clusters with the plumbing pre-solved. Reach for it when: you're fine-tuning or pretraining and want cluster ops handled.
GPU cloud
Affiliate
GPU cloud built on stranded/low-carbon energy; a data-center operator that rents compute. Reach for it when: you want capacity with an energy story, or you're watching the power layer.
GPU cloud
Affiliate
the second tier of dedicated GPU clouds competing on price and availability. Reach for it when: the big names are sold out or you're price-shopping a known workload.
GPU cloud
Affiliate
approachable GPU notebooks and machines under the DigitalOcean umbrella. Reach for it when: you want a friendly on-ramp — a notebook with a GPU attached, minutes from signup.
GPU cloud
Affiliate
hosted notebooks with free and paid GPU tiers; the classroom of the whole field. Reach for it when: learning, prototyping, or running someone's shared notebook.
Notebook + GPU
free GPU/TPU notebook hours plus the datasets and competitions community. Reach for it when: you want free compute for learning and a dataset already attached.
Notebook + GPU
Apple Silicon (local)
Affiliate
Macs with unified memory run surprisingly large open models via llama.cpp/MLX; the zero-marginal-cost inference path. Reach for it when: privacy, offline, or "I already own the hardware."
Local hardware
16 entries
Infrastructure
Where compute lives and how jobs get placed on it.
The hyperscalers
Affiliate
the largest cloud; AI surface spans Bedrock (managed model APIs), SageMaker (ML platform), and Trainium/Inferentia custom silicon. Reach for it when: your company is already on AWS — gravity wins — or you need Bedrock's model menu inside an enterprise-compliance envelope.
Cloud
Affiliate
the cloud wired most tightly to OpenAI; Azure OpenAI Service is how enterprises consume GPT models with enterprise controls. Reach for it when: you need OpenAI models with enterprise data guarantees, or Microsoft-stack integration.
Cloud
Affiliate
Vertex AI platform, Gemini APIs, and the only rentable TPUs; the most vertically integrated AI cloud. Reach for it when: you want Gemini at enterprise terms or TPU economics.
Cloud
Affiliate
the surprise fourth player; large GPU superclusters at aggressive prices, host to major AI-lab workloads. Reach for it when: negotiating big GPU commitments and you want a fourth bidder in the room.
Cloud
Serverless & job platforms (rent the abstraction, not the box)
Affiliate
serverless GPUs defined in Python; your function scales to hundreds of containers and to zero. Reach for it when: you want to ship a GPU workload without touching infrastructure — the fastest idea-to-production path on this shelf.
Serverless GPU
Affiliate
run and publish models behind a one-line API; the community catalog makes it a demo engine. Reach for it when: you want someone else's open model (image, video, audio, LLM) behind an API in one minute.
Model hosting
Affiliate
production-grade model serving with performance tooling (their Truss packaging) and autoscaling. Reach for it when: your own model needs real production serving with SLAs.
Model serving
Affiliate
inference on Cloudflare's edge network, priced per request, wired into their developer platform. Reach for it when: you're already on Workers and want small-model inference close to users.
Edge inference
Open source
open-source broker that runs your job on whichever cloud is cheapest/available, with spot-instance failover. Reach for it when: you have multi-cloud accounts and want the price arbitrage automated.
Scheduler (OSS)
Cluster software (the plumbing — reference for the dossier, and real choices for builders)
Open source
the container orchestrator that most GPU clouds speak natively; the default control plane of the modern cluster. Reach for it when: you operate services, not just jobs.
Orchestration (OSS)
Open source
the venerable HPC batch scheduler; still what most large training clusters actually run. Reach for it when: you're queueing multi-node training jobs on dedicated hardware.
Scheduler (OSS)
Affiliate · OSS
the distributed-Python framework used inside many training and serving stacks; Anyscale is its managed cloud. Reach for it when: scaling Python workloads across machines without writing distributed systems.
Distributed compute (OSS) / managed
Affiliate · OSS
containers; the unit of shipping in all of the above. Reach for it when: always — it's the envelope everything above expects.
Containers (OSS)
Training-run tooling (MLOps)
Affiliate
experiment tracking, model registry, and now LLM tracing (Weave); the lab notebook of ML teams. Reach for it when: any run you might need to compare against another run.
MLOps
Open source
the open-source experiment tracker and model registry; the self-hosted counterpart to W&B. Reach for it when: you want tracking without a vendor.
MLOps (OSS)
Affiliate
the GitHub of models: weights, datasets, Spaces demos, and the transformers library ecosystem. Reach for it when: finding, publishing, or evaluating any open model or dataset — the field's reference shelf.
Model hub
33 entries
Foundation
The models themselves: APIs, open weights, inference, fine-tuning, embeddings.
Frontier model APIs (closed weights, top capability)
Affiliate
the market-defining model line; broadest ecosystem, most third-party integrations. Reach for it when: you want the default the whole tool ecosystem assumes.
Model API
Affiliate
frontier models with a long-context, agentic-coding, and reliability reputation. Reach for it when: long documents, coding agents, and tasks where instruction-following under pressure matters.
Model API
Affiliate
frontier multimodal models with very large context windows and aggressive pricing tiers. Reach for it when: multimodal input, huge context, or price-performance at the mid-tier.
Model API
Affiliate
Europe's frontier lab; strong mid-size models, open and commercial, EU data residency. Reach for it when: efficiency-tier workloads or European compliance requirements.
Model API
Affiliate
enterprise-focused models (Command family) with strong embeddings and reranking; RAG-centric positioning. Reach for it when: enterprise search/RAG where embeddings + rerank + generation come from one vendor.
Model API
Affiliate
the Grok model family via API, with real-time X data as a differentiator. Reach for it when: you want frontier-class output wired to live social data.
Model API
Open-weight model families (reference — the weights are the product)
the open-weight family that reset the industry; the default base for fine-tunes and local deployment. Reach for it when: you need weights you control, with the biggest tooling ecosystem.
Open weights
Chinese lab whose open reasoning and MoE models matched closed frontier scores at a fraction of the claimed training cost. Reach for it when: open-weight reasoning capability per dollar is the criterion.
Open weights
the broadest open family — sizes from sub-1B to frontier-class, strong multilingual and coding variants. Reach for it when: you need a specific size/capability point; there's a Qwen for almost every slot.
Open weights
the efficient open line (and the Mixtral MoE lineage) from the Paris lab. Reach for it when: permissive-license weights at strong quality-per-parameter.
Open weights
Google's small open family, distilled from Gemini tech. Reach for it when: small, well-behaved models for on-device or cheap serving.
Open weights
OpenAI's return to open weights, released 2025 as gpt-oss-120b and gpt-oss-20b. Reach for it when: you want OpenAI-flavored behavior on your own hardware.
Open weights
Microsoft's family of small open-weight models trained on curated and synthetic data; they consistently outperform models larger than them on reasoning benchmarks. Reach for it when: you need a capable model that runs on a laptop, an edge device, or a tight inference budget.
Open weights
Inference providers (someone else's GPUs serving open models)
Affiliate
the leading open-model inference API; big catalog, fine-tuning, and dedicated endpoints. Reach for it when: you want open-weight models at production scale without owning serving.
Inference
Affiliate
speed-focused inference for open models, with function-calling and compound-AI features. Reach for it when: latency-sensitive production on open models.
Inference
Affiliate
the LPU silicon, consumed as an API; extreme tokens-per-second on supported models. Reach for it when: perceived speed is the feature — voice, live chat, rapid agent loops.
Inference
Affiliate
one API key, hundreds of models across every provider, with routing and price comparison built in. Reach for it when: you want to try everything without ten accounts, or route across providers by price/uptime.
Gateway / inference
Affiliate
the discount aisle of open-model inference; same models, thinner margins. Reach for it when: a batch workload where cost per million tokens is the only line that matters.
Inference
Affiliate
serverless inference across partner providers, integrated with the Hub. Reach for it when: you're already on the Hub and want one-click serving of what you find.
Inference
Self-hosted serving (your GPUs, open software)
Open source
the de facto open serving engine; continuous batching and PagedAttention made it the throughput standard. Reach for it when: serving an open model on your own GPUs in production.
Serving (OSS)
Open source
the fast-rising alternative serving engine, strong on structured output and multi-turn workloads. Reach for it when: you're chasing serving throughput and vLLM isn't winning your benchmark.
Serving (OSS)
Open source
one-command local model runtime for laptops and servers; the front door of local AI. Reach for it when: running open models locally with zero ceremony — development, privacy, or hobby hardware.
Local runtime (OSS)
Open source
the C++ inference engine that made CPU/consumer-GPU inference real; GGUF quantized models everywhere. Reach for it when: squeezing a model onto modest hardware, or embedding inference in an application.
Local runtime (OSS)
a polished desktop app for downloading and chatting with local models; llama.cpp with a friendly face. Reach for it when: you want local models without touching a terminal.
Local app
Fine-tuning & customization
Open source
open-source fine-tuning that cuts VRAM and time dramatically; the default answer to "how do I fine-tune on one GPU." Reach for it when: LoRA/QLoRA fine-tunes on a single consumer or rented GPU.
Fine-tuning (OSS)
Open source
config-driven open fine-tuning framework covering most model families and methods. Reach for it when: reproducible fine-tune recipes beyond a notebook.
Fine-tuning (OSS)
Affiliate
fine-tune small models on your production LLM traffic to replace expensive frontier calls. Reach for it when: a high-volume prompt is burning money on a frontier model and a distilled small model would do.
Fine-tuning SaaS
Affiliate
managed LoRA fine-tuning and serving with per-adapter economics (LoRAX lineage). Reach for it when: serving many fine-tuned variants cheaply.
Fine-tuning SaaS
Together / Fireworks fine-tuning
Affiliate
the inference providers' managed fine-tune paths. Reach for it when: you already serve there and want the weights close to the serving.
Fine-tuning SaaS
Embedding models (the quiet workhorses)
Affiliate
the default embedding API; cheap, good, everywhere. Reach for it when: starting any RAG build — it's the baseline to beat.
Embeddings
Affiliate
embedding-and-rerank specialist (now under MongoDB), strong domain-specific models (code, law, finance). Reach for it when: retrieval quality in a specific domain is the bottleneck.
Embeddings
Affiliate
multilingual embeddings and the most-used commercial reranker. Reach for it when: adding a rerank stage — often the single cheapest retrieval-quality win.
Embeddings
Open embeddings (nomic-embed, BGE/bge-m3, GTE family)
Open source
open-weight embedding models that run anywhere, including multilingual (bge-m3). Reach for it when: embeddings must run locally or for free — quality is close, price is zero.
Embeddings (OSS)
33 entries
Orchestration
Frameworks, agents, retrieval, vector stores, evals, gateways, guardrails.
App & agent frameworks
Affiliate · OSS
the biggest LLM-app ecosystem; LangGraph is its graph-based agent runtime, the current center of gravity. Reach for it when: building agent workflows with the largest integration catalog and hiring pool.
Framework (OSS)
Affiliate · OSS
the data-framework: ingestion, indexing, and retrieval pipelines over your documents. Reach for it when: the hard part of your app is the data side of RAG.
Framework (OSS)
Affiliate · OSS
role-based multi-agent framework ("crew of agents"), with a managed platform. Reach for it when: prototyping multi-agent workflows quickly with an opinionated structure.
Agents (OSS)
Open source
Microsoft's multi-agent research framework and its enterprise SDK sibling. Reach for it when: .NET/enterprise-Microsoft environments, or research-grade multi-agent patterns.
Agents (OSS)
Open source
OpenAI's first-party agent framework wired to its Responses API and tools. Reach for it when: all-in on OpenAI and want the paved road.
Agents (OSS)
Open source
Anthropic's agent harness — the machinery behind Claude Code, exposed as an SDK. Reach for it when: building long-running, tool-using agents on Claude.
Agents (OSS)
Open source
programs, not prompts: declarative pipelines whose prompts are optimized automatically against a metric. Reach for it when: you have an eval metric and want the prompt engineering done by optimization.
Framework (OSS)
Open source
the production-focused open RAG/pipeline framework from deepset. Reach for it when: enterprise search pipelines with a stability-first culture.
Framework (OSS)
the open standard for connecting tools and data to models; the USB port of the agent layer. Reach for it when: exposing any tool or data source to any MCP-speaking client — write once, plug everywhere.
Standard (open)
Vector databases & retrieval infrastructure
Open source
vector search inside the database you already run. Reach for it when: first choice for most teams — if your app has Postgres, start here and only leave when scale forces you.
Vector DB (OSS)
Affiliate
the managed vector-DB pioneer; serverless, zero-ops, scales without thought. Reach for it when: you want retrieval as a utility bill, not a system you operate.
Vector DB
Affiliate · OSS
open-source vector engine in Rust with strong filtering and a managed cloud. Reach for it when: self-host-first with a cloud escape hatch; heavy metadata filtering.
Vector DB (OSS)
Affiliate · OSS
open-source vector DB with hybrid (keyword+vector) search and built-in vectorizer modules. Reach for it when: hybrid search out of the box matters.
Vector DB (OSS)
Affiliate · OSS
the open vector DB built for billion-vector scale; Zilliz is its managed cloud. Reach for it when: the vector count has nine-plus digits.
Vector DB (OSS)
Open source
the developer-friendly embedded vector store; pip install and go. Reach for it when: prototypes, notebooks, and small apps — the SQLite of vector stores.
Vector DB (OSS)
Affiliate
object-storage-backed serverless vector search, cost-optimized for large, mostly-cold corpora. Reach for it when: huge, mostly-cold vector corpora where storage cost dominates.
Vector DB
Affiliate · OSS
the incumbent search engines, now with dense-vector support; where keyword search already lives. Reach for it when: you have an ES cluster and need hybrid retrieval without a new system.
Search (OSS)
Affiliate
vector search inside Atlas, next to your operational data. Reach for it when: your data is already in Mongo.
Vector DB
Evals, observability & tracing (the "does it actually work" layer)
Affiliate
LangChain's tracing/eval platform; the most common first observability tool. Reach for it when: you build on LangChain/LangGraph — the integration is free.
Evals/obs
Affiliate · OSS
open-source LLM engineering platform: tracing, evals, prompt management; self-host or cloud. Reach for it when: you want observability you can own — the open default.
Evals/obs (OSS)
Affiliate
eval-first platform for shipping AI products with regression-tested prompts. Reach for it when: evals are the center of your dev loop, not an afterthought.
Evals
Open source
open-source tracing and eval library from the ML-observability world. Reach for it when: OSS tracing with strong RAG-specific evals.
Evals/obs (OSS)
Affiliate
Weights & Biases' LLM tracing/eval layer. Reach for it when: your team already lives in W&B.
Evals/obs
Affiliate
proxy-based LLM observability — one base-URL change gives you logging, caching, and cost tracking. Reach for it when: you want cost visibility in five minutes.
Obs
Open source
open-source prompt testing and red-teaming from the CLI/CI. Reach for it when: prompt regression tests in CI, before anything reaches production.
Evals (OSS)
Gateways, routing & guardrails
Open source
open-source proxy that makes 100+ providers look like one OpenAI-compatible API, with keys, budgets, and fallbacks. Reach for it when: any team serving multiple models to multiple internal users — the standard self-hosted gateway.
Gateway (OSS)
Affiliate
managed AI gateway: routing, retries, guardrails, and cost controls as a service. Reach for it when: gateway-as-SaaS instead of running LiteLLM yourself.
Gateway
Open source
open framework for validating LLM inputs/outputs against schemas and policies. Reach for it when: structured-output enforcement and policy checks in code.
Guardrails (OSS)
Open source
open toolkit for conversational rails — topic bounds, jailbreak resistance, tool-use policies. Reach for it when: dialog systems that must stay on the rails.
Guardrails (OSS)
Affiliate
commercial prompt-injection and content-safety firewall API. Reach for it when: a production app faces untrusted user input and you want a managed shield.
Security
Automation platforms (agents for people who don't code)
Affiliate
the no-code automation incumbent, now with AI steps and agent features across thousands of app integrations. Reach for it when: wiring AI into business tools without engineering time.
Automation
Affiliate
visual automation with more control per dollar than Zapier. Reach for it when: complex branching flows on a budget.
Automation
Affiliate · OSS
source-available, self-hostable automation with first-class AI/LangChain nodes. Reach for it when: you want your automations (and their data) on your own box.
Automation (source-available)
54 entries
Application
What people actually subscribe to — assistants, coding, image, video, audio, work.
AI assistants (the front doors)
Affiliate
the product that started the consumer era; assistant, search, voice, agents, and an app ecosystem. Reach for it when: the general-purpose default; the one non-technical people already know.
Assistant
Affiliate
Anthropic's assistant; Projects, Artifacts, and a reputation for long-document work and writing quality. Reach for it when: serious documents, code, and long-context work.
Assistant
Affiliate
Google's assistant, woven through Workspace, Android, and Search. Reach for it when: you live in Google's ecosystem — it's increasingly just there.
Assistant
Affiliate
the AI answer engine: search that reads the web and cites its sources. Reach for it when: research questions where you want sources, not vibes.
Search assistant
Affiliate
one subscription, many models — a multi-model chat aggregator by Quora. Reach for it when: comparing frontier models side-by-side on one bill.
Aggregator
Affiliate
Microsoft's general-purpose AI assistant, built into Windows and Edge. Reach for it when: you are already in the Microsoft ecosystem and want an assistant that requires no extra setup.
Assistant
Affiliate
an AI research assistant that searches and synthesizes academic literature; extracts claims, methods, and findings from papers into structured tables. Reach for it when: you need to survey a research area and do not want to read 40 abstracts by hand.
Research
Affiliate
an answer engine that returns answers sourced and cited directly from peer-reviewed papers, not the open web. Reach for it when: you need a factual claim backed by published evidence, fast.
Research
a free scholarly search engine covering roughly 200 million papers, with AI-generated summaries and citation graphs. Reach for it when: you want deep literature coverage without a journal subscription or a paid tool.
Reach for it when: Research
free
Affiliate
a paid, ad-free search engine with built-in AI answers and no ad-based ranking; revenue comes from subscribers. Reach for it when: you are tired of search results shaped by advertisers and want to pay for quality instead.
Search
Coding (the killer app — deepest affiliate shelf on the site)
Affiliate
the most-deployed AI coding tool; completions, chat, and agent mode inside your editor. Reach for it when: the safe enterprise default in VS Code/JetBrains.
Coding
Affiliate
the AI-first code editor (a VS Code fork); repo-aware chat, multi-file edits, background agents. Reach for it when: you want the editor itself built around AI, not a plugin bolted on.
Coding
Affiliate
Anthropic's terminal/IDE coding agent — plans, edits across files, runs tests, ships. Reach for it when: agentic work on real codebases; the strongest pure-agent option.
Coding
Affiliate
AI-native editor known for its agentic "Cascade" flow. Reach for it when: Cursor-style editing with a different agent personality and price point.
Coding
Open source
open-source terminal pair-programmer that commits as it goes; bring your own API key. Reach for it when: CLI-native workflows with full control of model and cost.
Coding (OSS)
Open source
open-source agentic coding extensions for VS Code, BYO-key. Reach for it when: agent coding inside VS Code without a subscription.
Coding (OSS)
Affiliate
browser IDE where an agent builds and deploys whole apps from a prompt; hosting included. Reach for it when: idea-to-deployed-app with no local setup — the vibe-coding on-ramp.
Coding/platform
Affiliate
prompt-to-UI generator that emits production React/Tailwind, deployable on Vercel. Reach for it when: frontend scaffolding and marketing pages at speed.
Coding/UI
Affiliate
prompt-to-full-stack-app builder aimed at non-engineers. Reach for it when: a founder needs a working product without hiring yet.
App builder
Affiliate
StackBlitz's in-browser AI app builder; prompt, run, deploy in one tab. Reach for it when: fast prototypes with the runtime right there in the browser.
App builder
Affiliate
privacy-first completions that can run on-prem; trained to respect licensing. Reach for it when: enterprises whose counsel won't let code leave the building.
Coding
Affiliate
AI assistant across the IntelliJ family. Reach for it when: your team lives in JetBrains IDEs.
Coding
Image generation
Affiliate
the aesthetic benchmark of AI image generation; web editor and a famously opinionated look. Reach for it when: beauty is the deliverable — concept art, covers, mood.
Image
Affiliate
the strongest open-weight image family, from the original Stable Diffusion team; API and self-host. Reach for it when: open weights with near-frontier quality — control plus rights clarity.
Image (open + API)
Affiliate
the model line that made image generation open; the foundation of the entire local-image ecosystem. Reach for it when: the customization universe — LoRAs, ControlNet, fine-tunes.
Image (open)
Affiliate
image model known for actually rendering legible text. Reach for it when: posters, logos, anything with words in it.
Image
Affiliate
production-oriented image platform (Canva-owned) for game and marketing assets. Reach for it when: consistent asset pipelines, not one-off art.
Image
Affiliate
Adobe's commercially-safe image model inside Photoshop and Creative Cloud. Reach for it when: the legal department requires trained-on-licensed-content, or you live in Photoshop.
Image
Open source
the node-graph power tool for local image/video pipelines. Reach for it when: exact control over a generation pipeline, on your own GPU.
Image (OSS)
Affiliate
Canva's integrated AI suite for generating images, writing copy, and remixing designs inside the workspace non-designers already use. Reach for it when: your team needs polished visuals and no one on it is a designer.
Design
Affiliate
AI features built into Figma: generating first-draft layouts, filling content, and accelerating prototyping without leaving the design tool. Reach for it when: you are already designing in Figma and want AI to speed up the blank-canvas stage.
Design
Video generation
Affiliate
the AI-video pioneer (Gen-4 line) with editor tooling aimed at filmmakers. Reach for it when: image-to-video and video editing workflows with production intent.
Video
Affiliate
frontier text-to-video (with audio) via Gemini/Flow. Reach for it when: high realism and native soundtracks.
Video
Affiliate
OpenAI's video model and social-style app. Reach for it when: the ChatGPT-ecosystem video path.
Video
Affiliate
Kuaishou's video model; strong motion quality, aggressive pricing tiers. Reach for it when: high volume image-to-video on a budget.
Video
Affiliate
the fast-moving second rank of video generators, each with a distinct look. Reach for it when: shopping styles per shot — this category churns monthly.
Video
Affiliate
AI avatar video — talking presenters, dubbing, and translation. Reach for it when: training videos and localized talking-head content at scale.
Avatar video
Affiliate
an AI video platform that generates talking-head avatar videos from a script, no camera or studio needed. Reach for it when: you need a professional training video or corporate explainer and a filmed production is not in the budget.
Video
Audio, voice & music
Affiliate
the leading voice platform: TTS, cloning, dubbing, and conversational voice agents. Reach for it when: any production voice work — the quality bar the others chase.
Voice
Open source
the open speech-to-text model that made transcription free. Reach for it when: transcription on your own hardware at zero marginal cost.
STT (OSS)
Affiliate
production speech-to-text APIs with diarization, streaming, and audio intelligence. Reach for it when: transcription as a scaled product feature rather than a script.
STT
Affiliate
text-to-song generation — full tracks with vocals. Reach for it when: jingles, demos, and scored content without a studio.
Music
Affiliate
Suno's main rival; some listeners favor it for fidelity. Reach for it when: same job, different ear — try both.
Music
Affiliate
edit audio/video by editing the transcript; overdub voices, remove filler words. Reach for it when: podcast and video post-production without a timeline editor.
Audio/video editing
Writing, meetings & work
Affiliate
AI woven through the Notion workspace — writing, Q&A over your docs, meeting notes. Reach for it when: your team's knowledge already lives in Notion.
Productivity
Affiliate
the incumbent writing assistant, now generative. Reach for it when: org-wide writing polish with admin controls.
Writing
Affiliate
marketing-content platforms with brand-voice controls. Reach for it when: marketing teams producing on-brand copy at volume.
Marketing
Affiliate
prompt-to-presentation; decks, docs, and sites from an outline. Reach for it when: a presentable deck in ten minutes.
Presentations
Affiliate
meeting transcription and AI notes that join your calls. Reach for it when: meetings should write themselves up.
Meetings
Affiliate
search APIs built for AI agents — clean, structured web results for RAG and research loops. Reach for it when: your agent needs to search the web programmatically.
Search API
Affiliate
AI assistance woven across Word, Excel, PowerPoint, Outlook, and Teams for organizations on Microsoft 365. Reach for it when: your company is on M365 and you want AI where the work already happens, not in a separate tool.
Productivity
Google's research notebook that grounds every answer and audio overview in the specific sources you upload, not the open web. Reach for it when: you have a pile of documents and need to interrogate them without the model hallucinating outside them.
Reach for it when: Research
free
Affiliate
AI features native to Slack: thread summaries, channel recaps, and search over your workspace history. Reach for it when: your team runs on Slack and you need to catch up on a long thread without reading every message.
Productivity
Affiliate
a conversational data analysis tool: upload a spreadsheet or connect a database, ask questions, and get charts and statistics back. Reach for it when: you have data to explore and writing SQL or Python is not how you want to spend the afternoon.
Data
35 entries
Learn
Everything above churns; understanding compounds. Start here.
our open textbook: 6 parts, 34 chapters, plain English → math → runnable code, every mechanic bridged to its economics. Reach for it when: you want to understand the whole stack from a token upward — start here.
Reach for it when: Internal
free
the best free deep-learning course on the internet; build GPT from scratch, by hand, on video. Reach for it when: you're ready to actually build one, not read about one.
Reach for it when: Course
free
the visual intuition for neural nets, gradient descent, and transformers; unmatched animation-driven teaching. Reach for it when: before anything else — 90 minutes that make everything after easier.
Reach for it when: Video
free
Affiliate
Andrew Ng's course platform; short courses on RAG, agents, fine-tuning, usually co-taught with the tool vendors. Reach for it when: a structured 1-3 hour on-ramp to a specific technique.
Courses
the famous top-down free course: train real models first, theory later. Reach for it when: you learn by doing and want results in week one.
Reach for it when: Course
free
free hands-on courses (LLM, agents, deep RL) built around the HF ecosystem. Reach for it when: learning the open-source stack you'll actually use.
Reach for it when: Courses
free
the university courses on NLP and building LLMs, lectures free on YouTube. Reach for it when: academic depth with problem sets — the real thing.
Reach for it when: Course
free
Affiliate
the book version of Karpathy's arc: a working LLM, line by line. Reach for it when: you want the from-scratch path in book form.
Book
Affiliate
the production side: evals, RAG, fine-tuning decisions, serving — the field manual for shipping AI. Reach for it when: your questions changed from "how does it work" to "how do I ship it."
Book
the single most-cited visual explainer of the transformer. Reach for it when: attention finally needs to click.
Reach for it when: Article
free
the preprint server where the frontier appears first, before press releases. Reach for it when: reading the source, not the coverage.
Reach for it when: Papers
free
daily-curated trending papers with community discussion — a practical successor to Papers with Code (which wound down in 2025). Reach for it when: keeping up with what practitioners are actually reading.
Reach for it when: Papers
free
the sharpest practitioner blog in the field; every model release tested within hours, honestly. Reach for it when: a new model drops and you want a trustworthy hands-on read.
Reach for it when: Blog
free
the AI-engineering podcast and newsletter of record. Reach for it when: hour-deep interviews with the people building the tools above.
Reach for it when: Podcast
free
the two steadiest weekly digests: policy-aware and practitioner-plain respectively. Reach for it when: one email a week is your budget for keeping up.
Reach for it when: Newsletter
free
the best open analysis of model training, RLHF, and the open-vs-closed race. Reach for it when: you want to understand why labs make the choices they make.
Reach for it when: Newsletter
free/paid
the research group measuring compute trends, training costs, and scaling data — the field's quantitative memory. Reach for it when: any hard number about compute growth or model scale; we cite them ourselves.
Reach for it when: Research
free
Affiliate
Dylan Patel's semiconductor-and-datacenter research; the reference for the hardware layer's economics. Reach for it when: the Hardware and Infrastructure layers are your beat.
Reach for it when: Research
free/paid
Affiliate
the technical library subscription; every AI-engineering book above plus live courses. Reach for it when: your team expenses learning anyway.
Platform
free stats, probability, linear algebra, and calculus in short video units; the numeric literacy every model decision rests on. Reach for it when: your math is shaky and you want to close the gap before a course or book loses you.
Reach for it when: Course
free
the language's own tour, covering syntax to standard library with no assumptions. Reach for it when: you need to learn Python and want one authoritative source, not a patchwork of blog posts.
Reach for it when: Docs
free
the official guide to arrays, broadcasting, and vectorization; the substrate every ML library sits on. Reach for it when: you keep writing Python loops over data and suspect there is a faster way.
Reach for it when: Docs
free
Affiliate
the canonical intro course; the clearest explanation of how gradient descent, regression, and classification actually work. Reach for it when: you want a mental model of learning algorithms before you touch a framework.
Course
the readable bridge from statistics to machine learning, with labs in R and Python; free PDF from the authors. Reach for it when: you want to understand why a model behaves the way it does, not just how to run it.
Reach for it when: Book
free
the official tutorial sequence for the framework most research and production runs on; start at the 60-minute blitz. Reach for it when: you are ready to write training code and want to follow the path most engineers actually took.
Reach for it when: Docs
free
structured courses on prompt engineering, tool use, and MCP from the team that builds Claude. Reach for it when: you want authoritative guidance on getting the most out of Claude specifically.
Reach for it when: Course
free
copy-paste recipes for embeddings, function calling, structured output, and evals, maintained by the API team. Reach for it when: you know what you want to build but need a working code starting point.
Reach for it when: Code
free
an end-to-end production ML curriculum: design, testing, CI/CD, and serving, not just model training. Reach for it when: your model works in a notebook and you need to know what it takes to ship it.
Reach for it when: Course
free
the canonical enumeration of LLM application security risks: prompt injection, insecure output handling, data leakage, and more. Reach for it when: you are about to ship an LLM-powered feature and want to know what you need to harden.
Reach for it when: Reference
free
the federal reference standard for identifying, measuring, and governing AI risk; the vocabulary your legal and compliance teams will eventually use. Reach for it when: a stakeholder asks how you are managing AI risk and you need a recognized framework to point to.
Reach for it when: Reference
free
crowd-sourced human-preference rankings built from blind head-to-head model comparisons; the closest thing to a neutral scoreboard. Reach for it when: a new model is announced and you want to know where it actually ranks against the alternatives.
Reach for it when: Leaderboard
free
independent benchmarks measuring model quality, speed, and price side by side across providers. Reach for it when: you are choosing between API providers and need apples-to-apples numbers, not vendor claims.
Reach for it when: Leaderboard
free
the benchmark that evaluates coding agents on real GitHub issues from real open-source repositories. Reach for it when: someone claims their agent can code and you want to know what score that actually maps to.
Reach for it when: Benchmark
free
the annual, heavily-cited report compiling the field's key statistics on research output, adoption, policy, and capability. Reach for it when: you need citable numbers about where AI is and how fast it is moving.
Reach for it when: Report
free
the community of record for running open-weight models on your own hardware; dense with benchmarks, quantization guides, and early hands-on reports. Reach for it when: you want to run a model locally and need real-world setup experience, not documentation.
Reach for it when: Community
free