# Cyberax AI Playbook

A practical AI solutions catalog for any business. Maintained reference: what to use, where it runs, what it does well, and where it falls down. Each solution links to the models it references; each model links back to the solutions that use it.

Site: https://cyberax.com
Catalog: https://cyberax.com/ai-playbook
Models: https://cyberax.com/ai-playbook/models

## Communications & Customer Work

For support teams, sales teams, founders, and anyone whose day is shaped by inbound messages, calls, and meetings. AI that helps your team respond faster without sounding like AI.

- [AI agents for inbound qualification](https://cyberax.com/ai-playbook/ai-agents-inbound-qualification) — A chat-based AI assistant that pre-qualifies inbound leads — capturing role, company size, use case, and timing — before they reach a human sales rep. Without the friction of a 12-field form, and with the structured-capture discipline that stops the agent from inventing qualifying details the prospect never gave.
- [Auto-tag and route inbound social DMs](https://cyberax.com/ai-playbook/auto-route-social-dms) — A pipeline that watches your inbound DMs across LinkedIn, Twitter / X, Instagram, and TikTok — classifies each one by intent, routes the support cases to support and the press queries to press, and surfaces the partnership and exec asks to whoever should actually see them. So the founder stops reading every DM at 11pm trying to figure out which ones matter.
- [Detect churn signal from support patterns](https://cyberax.com/ai-playbook/churn-signal-from-support) — A pipeline that reads your support tickets continuously and surfaces the customers whose ticket pattern — frequency, tone, topic, escalation rate — predicts churn around 60 days out. Not "who's complaining today"; who is exhibiting the multi-month pattern that historically precedes cancellation.
- [Draft customer support replies that hold up to scrutiny](https://cyberax.com/ai-playbook/customer-support-reply-drafting) — **AI-assisted support reply drafting** means an AI model produces the first draft of each support response, grounded in your knowledge base, while a human agent reviews and sends. The workflow that keeps drafts in company voice, escalates the right tickets to humans, and stops hallucinated facts about your product from going out — including the confidence thresholds, the prompt shape, and the audit loop that keeps it honest.
- [Investor updates from BI data](https://cyberax.com/ai-playbook/investor-updates-from-data) — If you're a founder past seed stage and your monthly investor update keeps slipping, this is the pipeline that generates the draft from your BI dashboards rather than from a Sunday-night writing session. It pulls metrics, drafts the narrative, surfaces the right context per investor segment, and lands in your inbox Monday morning for editing.
- [Live-chat AI — when it works and when it actively hurts trust](https://cyberax.com/ai-playbook/live-chat-ai-when-to-use) — If you run customer experience or support, the question isn't "should we deploy AI in live chat?" It's "which conversations should the AI handle, and which should never reach it?" A framework for the conversations where AI helps, the ones where it damages trust, and how to route by conversation type rather than by category.
- [Meeting summaries people actually read](https://cyberax.com/ai-playbook/meeting-summaries-people-read) — If your team runs enough meetings that "remember what we agreed" has become a real cost, this is the workable system for turning recordings into short, scannable summaries with real action items. The tool choice, the prompt that matters, and the distribution step most teams skip.
- [Multilingual customer support routing](https://cyberax.com/ai-playbook/multilingual-customer-support) — If your support team is expanding internationally without a native speaker per market, this is the routing pattern that holds. Detect the inbound language, route to the right agent or pipeline, translate where appropriate, escalate where nuance matters — without the failure mode where machine-translated empathy lands wrong in a culturally-sensitive moment.
- [Outbound prospecting research at SDR scale](https://cyberax.com/ai-playbook/outbound-prospecting-research) — If your sales team sends 30+ outbound emails per rep per day, the math has shifted under you. This is the research pipeline that pulls company news, hiring signals, recent funding, and tech-stack changes per prospect, then generates first-touch outreach that actually references something specific. The architecture, the data sources, and the deliverability discipline that doesn't burn your sending domain.
- [Reactivation campaigns for dormant accounts](https://cyberax.com/ai-playbook/reactivation-campaigns-dormant-accounts) — If your dormant-account list is bigger than your active list, the generic "we miss you" blast is the worst version of the reactivation motion. This is the pipeline that segments by churn reason, references the specific cause in each message, and protects your sending domain along the way.
- [Reply templates learning from past conversations](https://cyberax.com/ai-playbook/reply-templates-from-past-replies) — If your support team has the same five questions arriving every week, this is the suggestion system that learns from the conversations you've already had — so similar inbound questions get answered with the same care without anyone re-typing the answer for the 50th time. Retrieval based on embeddings (the way AI represents meaning as numbers), context-aware suggestion, and a feedback loop that keeps the suggestions current.
- [Sales-call coaching at scale](https://cyberax.com/ai-playbook/sales-call-coaching) — Analyse every recorded sales call, surface the patterns that distinguish your top reps from the rest, and generate per-rep coaching reports their manager can actually use. The pipeline that turns "the manager listened to two calls this month" into systematic coaching, without making the recording process feel like surveillance.
- [Sales follow-up sequences with CRM context](https://cyberax.com/ai-playbook/sales-follow-up-sequences) — Generate personalised follow-up emails for every deal in your pipeline — using the CRM context the rep already has but doesn't have time to weave in — without the obvious AI-template tells that get replies marked as spam. The CRM-integration pattern, the brand-voice guardrails, and the deliverability hygiene that keeps you out of the promotions tab.
- [Slack channel summaries that catch what matters](https://cyberax.com/ai-playbook/slack-channel-summaries) — Turn a busy channel into a daily or weekly digest — decisions, action items, disagreements, with attribution — for distributed teams, async handoffs, and the post-vacation catch-up that otherwise takes a morning. The standing prompt, the threaded-reply handling, and the distribution that determines whether anyone reads it.
- [Summarize long email threads](https://cyberax.com/ai-playbook/summarize-long-email-threads) — Turn a 40-message thread into a brief a new joiner can read in 90 seconds — without losing the decisions, the disagreements, or the action items buried in the middle.
- [Transcribe audio at scale on a local machine](https://cyberax.com/ai-playbook/transcribe-audio-locally) — A self-hosted transcription pipeline that turns audio into text on your own GPU — no cloud API, no per-minute charges, no audio leaving the machine. With sizing notes for hardware, batching strategy, and the parts that bite teams in production.
- [Triage inbound email at scale](https://cyberax.com/ai-playbook/triage-inbound-email) — Turn a 200-message-a-day inbox into a sorted queue — urgency, topic, owner — without the model burying a real customer signal in the "FYI" bucket. The classification prompt, the routing rules, and the false-negative audit that keeps it honest.
- [Voice-of-customer reports from cross-channel feedback](https://cyberax.com/ai-playbook/voice-of-customer-reports) — Aggregate customer feedback from every channel — support tickets, app store reviews, NPS surveys, sales calls, social mentions — into themed reports product and leadership actually read. The cross-channel synthesis, the theme detection that doesn't just count keywords, and the trend tracking that catches emerging issues before they become quarterly board topics.
- [Voice transcription for sales calls and customer interviews](https://cyberax.com/ai-playbook/voice-transcription-for-sales-calls) — A practical setup for capturing, transcribing, and surfacing the signal from sales calls and customer interviews — the tool decision (revenue intelligence vs. lightweight notes), the privacy step everyone skips, and the workflow that turns transcripts into pipeline insight.

## Content & Marketing

For marketers, content teams, editors, and anyone producing words at any volume. AI assistance that doesn't strip the voice out of your writing.

- [Ad creative A/B testing at scale](https://cyberax.com/ai-playbook/ad-creative-ab-testing) — A workflow that generates twenty variants of each ad, tests them programmatically against the ad platforms, and lets performance data pick the winners. With the variant diversity that keeps the test results meaningful, instead of twenty rephrased versions of the same idea.
- [AI translation services compared](https://cyberax.com/ai-playbook/ai-translation-services-compared) — Five AI translation services and the human translator on retainer. What each is good for, what each costs per million characters, and the kind of content where AI translation will quietly betray you in ways a monolingual editor can't catch.
- [Brand-voice guardrails for marketing teams](https://cyberax.com/ai-playbook/brand-voice-guardrails) — A system that enforces your brand-voice rules across every piece of AI-generated content — captions, ads, social posts, emails, product copy — so the marketing team scales output without producing the "everyone sounds like ChatGPT" homogeneity that's now common at growing companies. The voice spec, the guardrail layer, and the monthly audit that catches drift.
- [Competitor monitoring with automated alerts](https://cyberax.com/ai-playbook/competitor-monitoring-automated) — A pipeline that watches your competitors continuously — pricing pages, product launches, hiring posts, marketing campaigns, social tone — and pings the right team when something material changes. So you stop discovering your top competitor's new pricing tier three months after their existing customers started asking your sales reps about it.
- [Content performance attribution](https://cyberax.com/ai-playbook/content-performance-attribution) — Content performance attribution is the practice of tracing which blog posts, whitepapers, and videos actually drive pipeline and deals — rather than judging content by page views. A pipeline that joins content engagement to CRM events, gives content fair credit across the customer journey, and tells you which categories are worth more investment.
- [Customer testimonial mining from reviews and support](https://cyberax.com/ai-playbook/customer-testimonial-mining) — **Testimonial mining** means scanning the places your customers already say nice things — G2 reviews, App Store ratings, support thank-yous, NPS comments, customer interviews — and pulling out quotes you can actually use in marketing, with the permissions sorted out. The pipeline that surfaces and operationalises the social proof you're already producing.
- [Generate alt text and image descriptions at scale](https://cyberax.com/ai-playbook/generate-alt-text-at-scale) — **Alt text** is a short written description of an image — read aloud by screen readers, shown when an image fails to load, and required for accessibility compliance. This is a practical pipeline for generating it across hundreds or thousands of images, with the prompt that produces good descriptions, the editorial pass that keeps them honest, and the empty-alt rule most pipelines get wrong.
- [Generate FAQ content from existing docs](https://cyberax.com/ai-playbook/generate-faq-from-docs) — A workflow for building an **FAQ** — frequently asked questions — that reflects what your users actually ask, not what a marketer guessed at three years ago. AI clusters real questions from your support tickets, KB articles, and chat logs, then writes the answers from your existing docs. Includes the clustering pass, the grounding rule, and the monthly refresh that keeps it useful.
- [Hook generation for short-form video](https://cyberax.com/ai-playbook/hook-generation-short-form-video) — A **hook** is the opening line of a short-form video — the first three seconds that decide whether the viewer keeps watching or scrolls on. This piece is the workflow that generates 30 hook variants for one topic, scores them against the patterns that survive the algorithm's scroll-stop test, A/B tests the survivors, and builds a library of winners. Includes the patterns that consistently work, the ones that don't, and why "Hey guys, today I'm going to talk about…" is the most expensive opener on the platform.
- [First-draft marketing copy without the AI tells](https://cyberax.com/ai-playbook/marketing-copy-without-ai-tells) — If you're an in-house marketer using AI for first drafts, the problem isn't speed — it's that the output reads as AI-written. A repeatable workflow for generating first-draft marketing copy with an LLM (the technology behind ChatGPT, Claude, and Gemini) that doesn't sound like one. Voice samples, prompt shape, the surgical edit pass, and the standing list of AI tells to strip on every pass.
- [Newsletter automation from weekly content](https://cyberax.com/ai-playbook/newsletter-automation) — If your weekly newsletter has slipped to fortnightly and then monthly, the bottleneck is the assembly, not the writing. This pipeline auto-aggregates your week's content from Slack, blog posts, social, customer wins, and product changelogs — then drafts the newsletter in your established voice for editing rather than writing from scratch.
- [Personalised email at 1-to-1000 scale](https://cyberax.com/ai-playbook/personalised-email-at-scale) — If you run lifecycle or growth marketing and your sends are dominated by one-message-to-all blasts, this is the pipeline for genuinely-different emails per recipient. Use customer data, behaviour signals, and segment context to vary the body, the offer, and the call-to-action per email. Not merge-fields — genuinely different messages that share a campaign goal.
- [Programmatic SEO at scale](https://cyberax.com/ai-playbook/programmatic-seo-at-scale) — If your SEO strategy depends on long-tail organic traffic, this is how to generate thousands of city, product, comparison, and use-case pages without triggering Google's thin-content penalty. The data sources, the page-template architecture, the quality gate that distinguishes useful programmatic from spam, and the index management that keeps the good pages and unpublishes the bad.
- [Prompt engineering patterns for content teams](https://cyberax.com/ai-playbook/prompt-engineering-patterns-content-teams) — If you run an in-house content team and your AI drafts read as generic, the fix isn't a smarter model — it's a better prompt. Five reusable patterns that take output from generic to on-brand: role priming, context packing, format constraints, voice training, and iteration scaffolds. With examples, when each fails, and how to compose them.
- [Repurpose a podcast episode into 5 written pieces](https://cyberax.com/ai-playbook/repurpose-podcast-episode-into-pieces) — If you produce a podcast, long-form interview, or webinar, this is the practical pipeline for turning one hour of recording into a blog post, a newsletter, social pull-quotes, short-form clips, and an FAQ section — without making each one feel like leftovers.
- [SEO content audit at scale](https://cyberax.com/ai-playbook/seo-content-audit) — Audit thousands of existing pages — for quality, freshness, performance, cannibalisation — and produce the prioritised list of pages to refresh, consolidate, or unpublish. Not a one-time spreadsheet; a continuous pipeline that catches content rot before it drags your domain's rankings down.
- [Social listening and brand-mention triage](https://cyberax.com/ai-playbook/social-listening-brand-mentions) — A pipeline that catches every mention of your brand across social platforms, classifies them by intent (compliment, complaint, journalist query, customer support, partnership feeler), and routes them to the team that can actually respond — so your founder isn't doom-scrolling Twitter at midnight trying to remember which mentions need a reply.

## Operations & Knowledge

For ops teams, internal tools owners, and anyone who needs to extract, organize, or search information that lives across files, archives, or systems.

- [Audit-trail generation from system logs](https://cyberax.com/ai-playbook/audit-trail-from-logs) — A pipeline that turns raw system logs (auth events, access records, change history) into the audit narrative your compliance team can read — not a 50,000-line CSV nobody opens. AI summarisation, anomaly detection, and the workflow that satisfies SOC 2 / ISO 27001 / HIPAA auditors without an eight-week prep cycle.
- [Auto-categorize support tickets by topic and urgency](https://cyberax.com/ai-playbook/auto-categorize-support-tickets) — A workflow that tags inbound tickets reliably enough to route them automatically — combining classical machine learning (cheap, fast, explainable) and LLMs (flexible but expensive). With confidence thresholds, a human-review queue for the ambiguous cases, and the evaluation loop that keeps the system honest as your product evolves.
- [Auto-generate documentation from PRs and code](https://cyberax.com/ai-playbook/auto-generate-docs-from-prs) — A continuous-integration pipeline that keeps your developer docs in sync with the codebase — drafting doc updates the moment a pull request (PR) merges, routing them to the docs owner for review, and stopping the gap between code and docs from growing into the year-long backlog you can't recover from.
- [Automated invoice and receipt processing](https://cyberax.com/ai-playbook/automated-invoice-processing) — Get invoice and receipt data — vendor, amount, line items, dates, tax — out of PDFs and into your accounting system automatically, without someone keying numbers in by hand at 11pm. The approach that actually works on real vendor invoices, the checks that catch silent mistakes, and the human-review queue that handles the tricky cases.
- [Build a private knowledge base your team can search](https://cyberax.com/ai-playbook/build-a-private-knowledge-base) — A practical setup for "chat with our docs" — a system that lets your team ask plain-language questions across your internal documents and get grounded answers with citations. The framework choice, the cheap-vs-managed vector database call (a database that stores meaning-as-numbers so you can search by similarity), and the hybrid-search-plus-rerank pattern the field has converged on. With cost ranges and the parts that bite teams in production.
- [Compliance evidence collection for SOC 2 / ISO 27001](https://cyberax.com/ai-playbook/compliance-evidence-collection) — A pipeline that pulls the screenshots, configurations, access logs, and policy snippets your auditor wants — automatically, on a schedule, organised by control. Turns "SOC 2 evidence collection week" into a continuous background process, and makes the auditor's questions take minutes instead of days.
- [Contract review and clause extraction](https://cyberax.com/ai-playbook/contract-review-clause-extraction) — **Clause extraction** means pulling out the parts of a contract that matter — termination, liability, IP, renewal, governing law — and comparing each one to the terms your company normally accepts. Not legal advice, not legal review. A first-pass triage so your lawyer spends an hour on the contract that needs an hour, not an hour reading every routine NDA.
- [CRM data hygiene at scale](https://cyberax.com/ai-playbook/crm-data-hygiene) — **CRM data hygiene** means keeping the records in your sales database accurate, consistent, and free of duplicates. This page is the pipeline that finds duplicate accounts, fills in missing fields from public sources, fixes inconsistent inputs, and flags the records that genuinely need a human's eyes — without an offshore data team typing for six months.
- [Customer health scoring from product and support signals](https://cyberax.com/ai-playbook/customer-health-scoring) — **Customer health scoring** is a single number per account that tells you which customers are healthy, which are drifting, and which are at real risk of cancelling. It's built from product usage, support interactions, billing events, and the softer signals you'd otherwise have to read every conversation to catch. The page covers the model that holds up across customer segments and the workflow that turns the score into actual renewal motion.
- [Data entry automation from scanned forms](https://cyberax.com/ai-playbook/data-entry-from-scanned-forms) — Turn a backlog of paper forms — intake forms, applications, surveys, handwritten records — into structured rows in your CRM or database. **OCR** (optical character recognition — software that reads text from images and PDFs) handles the reading, an LLM (the technology behind ChatGPT, Claude, and Gemini) handles the structured extraction, and validation rules catch the inevitable misreads. The pipeline that's worth building when the alternative is two full-time staff typing for the next year.
- [Document AI services compared](https://cyberax.com/ai-playbook/document-ai-services-compared) — A **document AI service** reads invoices, contracts, forms, and other documents — pulling out the fields you need so the data can flow into a database or accounting system. This piece compares AWS Textract, Google Document AI, OpenAI Vision, Azure Document Intelligence, and a do-it-yourself pipeline with open-weights models — what each is good for at scale, what each costs per page, and where each breaks on real documents.
- [Document classification at scale](https://cyberax.com/ai-playbook/document-classification) — **Document classification** is the practice of assigning each file in an archive a consistent label — topic, department, sensitivity, doc-type — so you can route it, search it, or apply a retention policy. This piece uses embeddings (the way AI represents meaning as numbers) for the obvious cases and an LLM (the technology behind ChatGPT, Claude, and Gemini) only for the residual that needs reasoning. Includes the taxonomy that holds up, the routing layer that doesn't break under change, and the audit that catches the silent failures.
- [Email-to-task automation](https://cyberax.com/ai-playbook/email-to-task-automation) — **Email-to-task automation** turns every 'can you send the deck by Friday' or 'please review the proposal' hiding in your inbox into a structured ticket — with owner, due date, and a link back to the source email. The extraction prompt, the 'is this actually a task for me?' gate that stops false positives, and the close-the-loop hook most pipelines skip.
- [Expense report categorization and anomaly detection](https://cyberax.com/ai-playbook/expense-report-anomaly-detection) — **Anomaly detection** for expenses means a system that quietly sorts every submission into a category, checks each one against your policy, and only escalates the few that look unusual. The classification that holds, the signals that catch what humans miss, and the policy-violation flags that don't read as accusatory when they fire on the CEO.
- [Extract structured data from PDFs at scale](https://cyberax.com/ai-playbook/extract-structured-data-from-pdfs) — **Structured data extraction** means turning a PDF — an invoice, a contract, a form, a scanned document — into named fields your downstream system can use (vendor, total, date, line items). This is the pipeline, and the LLM-vs-OCR decision that determines whether your project ships in a week or stalls for a quarter.
- [Federated search across your tools](https://cyberax.com/ai-playbook/federated-search-across-tools) — **Federated search** means one search bar that queries every tool you use — Slack, Drive, Notion, Linear, Confluence, Gmail — and ranks the combined results in a single list, with each result labelled by where it came from. The architecture, the per-tool connector quirks, the access-control layer that respects each tool's permissions, and the freshness signal that keeps results trustworthy.
- [Find patterns in customer feedback](https://cyberax.com/ai-playbook/find-patterns-in-customer-feedback) — **Pattern-finding in feedback** means letting AI cluster every customer message you receive — support tickets, NPS comments, sales calls, app store reviews — by what it's about, so the recurring themes surface on their own. A practical pipeline that turns the pile into a short list of priorities, without false confidence and without losing the long tail.
- [Insurance policy portfolio review](https://cyberax.com/ai-playbook/insurance-policy-portfolio-review) — If you handle insurance for a growing company, the annual renewal cycle catches most teams flat-footed. This pipeline extracts coverage details from every active policy — limits, deductibles, exclusions, named insureds, renewal dates — and surfaces gaps and overlaps before your broker sends you a proposal you don't have the context to evaluate.
- [Internal Q&A bot over company docs](https://cyberax.com/ai-playbook/internal-qa-bot-company-docs) — If your people-ops or IT team is fielding the same questions every week, this is the bot that learns from your wiki, your handbooks, and your private docs — and answers "what's our policy on X" without inventing policy or leaking a draft that wasn't supposed to ship. The retrieval that grounds answers in real sources, the fact-verification that catches wrong-document failures, and the access controls most teams skip.
- [Lease and vendor renewal tracking](https://cyberax.com/ai-playbook/lease-vendor-renewal-tracking) — If you handle vendor contracts and your portfolio is growing, the spreadsheet you use to track renewals is about to fail you. This pipeline extracts renewal dates and notice windows from every contract, surfaces the 60-day windows weeks before action is needed, and routes the decision to the right owner — finance, legal, or department head — with the contract attached.
- [Onboarding documentation generation for new hires](https://cyberax.com/ai-playbook/onboarding-doc-generation) — If your people-ops team is writing custom onboarding plans for every new hire, you've outgrown that workflow. This pipeline auto-generates role-specific plans — week-one priorities, key contacts, system-access checklist, SOP (standard operating procedure) links — by pulling from your existing docs, tailoring per role, and staying current as the org changes.
- [Build a private RAG with no third-party calls](https://cyberax.com/ai-playbook/private-rag-no-third-party-calls) — If you're an engineer at a security-sensitive or regulated company, this is a retrieval-augmented setup — giving the AI access to your specific documents — that never leaves your own machines. Embeddings, vector store, retrieval, and response synthesis all run locally. Hardware sizing, latency benchmarks, ongoing eval, and the failure modes specific to this configuration.
- [Procurement RFP response comparison](https://cyberax.com/ai-playbook/procurement-rfp-comparison) — If you're a procurement or finance lead evaluating multiple vendor RFP (request-for-proposal) responses, you can stop spending a week reading them in parallel. Extract pricing, capabilities, terms, and reference data into a structured matrix, flag the responses that materially diverge, and surface the trade-offs the marketing pages obscure.
- [Quote-to-cash automation](https://cyberax.com/ai-playbook/quote-to-cash-automation) — If you run RevOps or finance at a growing B2B company, this is the workflow from approved deal to invoiced revenue — pulling deal terms from the CRM, generating the contract, routing for signature, creating the invoice, and reconciling payment — without the manual handoffs that cost finance and ops days per deal. The integration layer, the approval gates, and the audit trail that holds up.
- [Resume screening with anti-bias guardrails](https://cyberax.com/ai-playbook/resume-screening-with-guardrails) — A first-pass resume screener that ranks candidates against job criteria, not against patterns that act as proxies for race, gender, school prestige, or age. The structured extraction, the bias-aware scoring, and the audit trail that holds up when legal asks how the screen worked.
- [SOP extraction from interviews](https://cyberax.com/ai-playbook/sop-extraction-from-interviews) — Turn the "how I actually do this" interview with a senior operator into a written, structured standard operating procedure — without the operator writing it themselves and without the documentation gap that closes when they leave the company. The interview structure that captures real workflow, the extraction pipeline that produces something a new hire can follow, and the gap-finding pass that surfaces the steps the operator forgot to mention.
- [Tax document classification and extraction](https://cyberax.com/ai-playbook/tax-document-extraction) — Automate the front-end of tax-document processing — sort W-2s, 1099s, receipts, invoices, brokerage statements, K-1s, charitable donation records — into the right buckets, extract the line items, and prep the data for review. Without the manual sorting that consumes a CPA's time before the actual tax work begins.
- [Vector databases compared](https://cyberax.com/ai-playbook/vector-databases-compared) — pgvector, Pinecone, Qdrant, Weaviate, Milvus — what each vector database is good for, what each costs, and the honest default for teams that haven't outgrown their existing Postgres.
- [Vendor risk assessment from questionnaires](https://cyberax.com/ai-playbook/vendor-risk-assessment) — Process the SOC 2 reports, security questionnaires, DPAs, and SIG forms that vendors send during procurement — extract the meaningful risk signals, compare against your security baseline, and flag the gaps that need follow-up. Without a security analyst reading every 200-question questionnaire end-to-end.

## Tool Decisions

Side-by-side evaluations and decision frameworks. Not "which is best" — "which is best for what." Updated when the answers change.

- [AI coding tools for non-engineers](https://cyberax.com/ai-playbook/ai-coding-tools-for-non-engineers) — For founders, ops leads, marketers, and analysts who need to ship code without a computer-science degree. Which AI tools are safe to learn on, which produce code you can hand off later, and where the line sits between "AI helped me build this" and "AI built something that will haunt me."
- [AI meeting assistants compared (Otter, Fireflies, Granola, Read AI)](https://cyberax.com/ai-playbook/ai-meeting-assistants-compared) — Five AI meeting tools that record your calls, transcribe them, and produce a summary you can paste into a follow-up — and that diverge sharply on output quality, search, integrations, and what happens to your recordings after. Where Otter, Fireflies, Granola, Read AI, and Fathom each fit, and the decision rules per team type.
- [AI search APIs compared (Perplexity, Tavily, SerpAPI + LLM)](https://cyberax.com/ai-playbook/ai-search-apis-compared) — Five APIs (application programming interfaces — the way one piece of software calls another) that give an AI workflow real-time web search. Perplexity, Tavily, SerpAPI plus your own model, Brave, and Exa each take a different approach. Where each fits, the latency and accuracy trade-offs, and the integration cost most pages don't show.
- [AI video editing tools compared (Descript, Captions, Opus Clip)](https://cyberax.com/ai-playbook/ai-video-editing-tools-compared) — Four AI-augmented video tools that solve different parts of the production pipeline. Where Descript's text-based editing earns its place, where Captions wins at vertical-format social, where Opus Clip dominates the long-form-to-short-form workflow, and how Adobe Premiere's AI features fit traditional editors.
- [AI writing tools compared](https://cyberax.com/ai-playbook/ai-writing-tools-compared) — Five AI writing platforms compared — Jasper, Copy.ai, Writer, ChatGPT Team, and Claude Team. What each is genuinely good at, what each costs per seat, and the honest case for skipping the wrapper tools entirely and using raw GPT or Claude with a shared prompt library.
- [ChatGPT Team vs Claude Pro vs Microsoft Copilot for small business](https://cyberax.com/ai-playbook/chatgpt-vs-claude-vs-copilot-business) — Three flagship AI subscriptions that look similar on the marketing pages and actually serve different teams well. Where ChatGPT Team's broader ecosystem pays off, where Claude's long-context work (handling book-length documents in one prompt) and stronger coding shines, and where Microsoft 365 Copilot's deep Office integration earns its price.
- [Cursor vs Copilot vs Claude Code for coding assistance](https://cyberax.com/ai-playbook/cursor-vs-copilot-vs-claude-code) — An **AI coding assistant** is a tool that helps developers write, edit, and debug code by suggesting lines, refactoring files, or completing whole tasks on its own. This is a side-by-side of the three most teams are choosing between in May 2026 — what each is genuinely best at, where each falls down, and the hybrid setups working developers actually use.
- [Embeddings models compared for semantic search](https://cyberax.com/ai-playbook/embeddings-models-compared) — An **embedding** is the way AI represents the meaning of a piece of text as a list of numbers, so similar things can be compared mathematically. An **embedding model** is the model that produces those numbers. This page compares five model lines — OpenAI text-embedding-3, Cohere embed v3, Voyage AI, Jina, and the open-source sentence-transformers family — on retrieval quality, dimension-and-cost trade-offs, and how to test them on your actual data before committing.
- [GPT vs Claude vs Gemini for business writing](https://cyberax.com/ai-playbook/gpt-vs-claude-vs-gemini-business-writing) — A practical side-by-side comparison of the three flagship LLMs — large language models, the technology behind ChatGPT, Claude, and Gemini — for business-writing work in May 2026. Voice, instruction-following, edit quality, pricing, and the picks for specific use cases.
- [Image generation models for business use](https://cyberax.com/ai-playbook/image-models-for-business-use) — If your team needs AI-generated images for marketing, five tools dominate — Flux, Midjourney, SDXL, Imagen, and gpt-image-2. They produce different output, cost different amounts, and carry sharply different licensing terms. Here's where each one wins for real commercial work.
- [When to run AI locally vs in the cloud](https://cyberax.com/ai-playbook/local-vs-cloud-ai) — If you're an engineering leader or founder whose AI bill is starting to matter, the "self-host or use the cloud?" decision rarely has one right answer. A practical framework with break-even math, the categories of workload that belong on each side, and the hybrid pattern most teams land on in 2026.
- [n8n vs Zapier vs Make for AI automations](https://cyberax.com/ai-playbook/n8n-vs-zapier-vs-make) — Three platforms that connect your business apps together and run AI tasks automatically — Zapier, Make, and n8n. Where each one fits which kind of team, what they actually cost as your volume grows, and how the choice between cloud-hosted and self-hosted shapes the day-to-day work of running automations.
- [No-code AI app builders compared](https://cyberax.com/ai-playbook/no-code-ai-builders-compared) — If you're a non-engineer trying to build an MVP, or an engineer in a hurry, five tools dominate — Bolt, Lovable, v0, Replit Agent, and Cursor. Where each one fits, where the prototype-to-production cliff lives, and what "no-code" actually means when the code still ships and still needs maintenance.
- [ElevenLabs vs Murf vs Play.ht for voice generation](https://cyberax.com/ai-playbook/voice-generation-tools-compared) — Three voice-generation services that look similar on the demo page and diverge sharply on production use. Where ElevenLabs' quality earns the premium, where Murf's structured workflow fits enterprise, where Play.ht's ecosystem makes sense for some teams — with honest licensing and clone-voice considerations.
- [Whisper API vs Deepgram vs AssemblyAI](https://cyberax.com/ai-playbook/whisper-vs-deepgram-vs-assemblyai) — Three transcription APIs that look interchangeable on the marketing page and diverge sharply on what they're actually good at. Where each one wins, where each one quietly loses, and the decision rule that fits your audio.

## Foundations

For readers without a technical background trying to understand what AI is doing under the hood. No code, no jargon, no acronym soup.

- [AI hallucinations explained](https://cyberax.com/ai-playbook/ai-hallucinations-explained) — Why LLMs — the technology behind ChatGPT, Claude, and Gemini — confidently produce wrong answers, how often it happens in 2026 (the honest answer is "more than vendors imply"), and the four mitigations that actually move the needle in production systems.
- [AI privacy — what to watch for](https://cyberax.com/ai-playbook/ai-privacy-what-to-watch-for) — An operator's checklist for the AI-vendor data questions that actually matter — what gets used to train future models, who retains what for how long, where regulations bite, and the policy changes that quietly flipped consumer defaults in 2025.
- [AI procurement checklist for non-technical buyers](https://cyberax.com/ai-playbook/ai-procurement-checklist) — A 20-question checklist for evaluating an AI vendor before you sign — the questions that actually predict deployment success, the red flags that should kill a deal, and the contract terms that matter more than the demo. In plain language a non-technical buyer can run without a CTO in every meeting.
- [AI risk assessment for legal and compliance teams](https://cyberax.com/ai-playbook/ai-risk-for-legal-compliance) — A framework for evaluating AI deployments against the legal and compliance surface that actually applies to your business — data privacy, regulatory exposure, IP and copyright, bias and discrimination, contractual obligations. Produces operational "approve with these controls" decisions instead of the over-broad "AI is risky everywhere" framing.
- [AI security risks for businesses](https://cyberax.com/ai-playbook/ai-security-risks-for-businesses) — The security risks specific to AI deployments — prompt injection, model exfiltration, training-data poisoning, supply-chain attacks on AI components, employees using consumer AI for confidential work — that the standard security playbook doesn't cover. What's actually exploitable today, what's theoretical, and which controls matter for which business shape.
- [Vendor lock-in risks with AI](https://cyberax.com/ai-playbook/ai-vendor-lock-in-risks) — Every AI product you wire into your operation creates a small dependency on the vendor; the cumulative dependency over 18 months becomes a strategic exposure. Where lock-in lives in modern AI stacks, the realistic mitigation patterns, and which lock-in is acceptable vs unacceptable for which workloads.
- [AI vs ML vs deep learning vs LLMs](https://cyberax.com/ai-playbook/ai-vs-ml-vs-deep-learning-vs-llms) — The family tree of four terms that get used interchangeably in business meetings — artificial intelligence, machine learning, deep learning, and large language models. What each one actually means, where each fits in your operations, and which one solves your particular problem.
- [Data leakage in AI tools — what to watch for](https://cyberax.com/ai-playbook/data-leakage-in-ai-tools) — **Data leakage** is when information you send to an AI tool ends up somewhere you didn't intend — used to train the vendor's next model, retained on their servers, exposed to other customers, or surfaced in a public search index. This page walks through the specific paths data can leak, what the vendors' marketing pages downplay, and which deployments are safe versus which need redesign.
- [The economics of self-hosting AI](https://cyberax.com/ai-playbook/economics-of-self-hosting-ai) — **Self-hosting AI** means running models on hardware you own (or rent), instead of paying a vendor like OpenAI or Anthropic per request through their API (the way one piece of software calls another). This page covers when the API math stops working and self-hosting starts paying off — with honest numbers on hardware, electricity, ops cost, and the engineering investment most teams underestimate.
- [Embeddings explained without math](https://cyberax.com/ai-playbook/embeddings-explained-without-math) — How AI systems turn text into positions on an invisible map of meaning — what they're used for, when they fail, and what you actually need to know to make decisions about them.
- [Fine-tuning vs RAG vs prompt engineering: which when](https://cyberax.com/ai-playbook/fine-tuning-vs-rag-vs-prompt-engineering) — Three ways to change what an AI model does for you. **Prompt engineering** is rewriting your instructions until you get the output you want. **RAG** (retrieval-augmented generation) gives the AI access to your specific documents so it answers from your information. **Fine-tuning** means training a model further on your specific data. Climbed in the wrong order, they get expensive fast. A decision framework for the operator who's been told "we should fine-tune."
- [The hidden costs of "free" AI tools](https://cyberax.com/ai-playbook/hidden-costs-of-free-ai-tools) — A **free AI tool** is one you don't pay a subscription for — ChatGPT free, Claude free, Gemini free, or an open-weights model running on someone else's server. Each one recovers its costs in ways that don't show up on an invoice — data, switching cost, time tax, and quality cap. A plain-language framework for spotting what you're actually paying.
- [Open-source vs proprietary AI — practical tradeoffs](https://cyberax.com/ai-playbook/open-source-vs-proprietary-ai) — If you're choosing AI infrastructure for a workload that actually matters, the "open vs closed" debate is the wrong frame. The real questions are capability, control, cost at scale, and vendor risk. For most teams the right answer is a hybrid — each model serves the workload it's best at. Where each side wins, where each side loses, and the decision framework that beats religious-war framing.
- [RAG explained without acronyms](https://cyberax.com/ai-playbook/rag-explained-without-acronyms) — If you're an operator or founder considering a "chat with our docs" tool, this is what retrieval-augmented generation actually is, when you genuinely need it, and what it costs to set up. Written for the person making the decision — not the engineer wiring it up.
- [ROI of AI projects — a realistic framework](https://cyberax.com/ai-playbook/roi-of-ai-projects) — A framework for honestly measuring AI return on investment across four categories — cost reduction, capacity creation, revenue enablement, and strategic optionality. Steers between the credulous numbers in vendor decks and the cynical numbers that ignore everything except direct labour savings.
- [Tokens, context windows, and what they cost](https://cyberax.com/ai-playbook/tokens-context-windows-and-cost) — The bill of materials for any AI feature — what a token is, what a context window does, how the input-vs-output pricing works, and the worked examples that turn vendor pricing pages into a sensible budget.
- [What "AI agents" actually are (and aren't)](https://cyberax.com/ai-playbook/what-ai-agents-actually-are) — The "AI agent" pitch is now applied to everything from a smart chatbot to a multi-step workflow tool to true autonomous software. What the term actually means in 2026, where the capability lives today, and how to evaluate the gap between agent demos and production-ready agents in the work you're considering buying.
- [What an LLM actually does for a business](https://cyberax.com/ai-playbook/what-an-llm-does-for-a-business) — What large language models — the technology behind ChatGPT, Claude, and Gemini — really are, what they're genuinely useful for, and where they fail. In plain language, with no jargon, no hype, and no acronyms left undefined.
- [When AI is the wrong tool](https://cyberax.com/ai-playbook/when-ai-is-the-wrong-tool) — A clear-eyed catalogue of the categories where AI underperforms simpler approaches — what to use instead, and why most "AI failure" stories are actually "AI used for the wrong thing" stories.
- [Why most "AI strategies" fail in the first 90 days](https://cyberax.com/ai-playbook/why-ai-strategies-fail) — The first-quarter failure modes are remarkably consistent across companies — over-broad scope, lack of operational ownership, misalignment between leadership pitch and team execution, and the gap between AI demos and production reality. What the pattern looks like, what the operational fix is, and how to set up an AI program that survives past its honeymoon.

## Browse by category

Each category has its own landing page with the full list and client-side filters by type, difficulty, and hosting.

- [Communications & Customer Work](https://cyberax.com/ai-playbook/category/communications) — 19 solutions. For support teams, sales teams, founders, and anyone whose day is shaped by inbound messages, calls, and meetings. AI that helps your team respond faster without sounding like AI.
- [Content & Marketing](https://cyberax.com/ai-playbook/category/content-marketing) — 17 solutions. For marketers, content teams, editors, and anyone producing words at any volume. AI assistance that doesn't strip the voice out of your writing.
- [Operations & Knowledge](https://cyberax.com/ai-playbook/category/operations-knowledge) — 29 solutions. For ops teams, internal tools owners, and anyone who needs to extract, organize, or search information that lives across files, archives, or systems.
- [Tool Decisions](https://cyberax.com/ai-playbook/category/tool-decisions) — 15 solutions. Side-by-side evaluations and decision frameworks. Not "which is best" — "which is best for what." Updated when the answers change.
- [Foundations](https://cyberax.com/ai-playbook/category/foundations) — 20 solutions. For readers without a technical background trying to understand what AI is doing under the hood. No code, no jargon, no acronym soup.

## Models

Reference catalog of AI models. Each entry includes modality, license, where to try the model, and the playbook solutions that reference it.

- [GPT-5.6 Luna](https://cyberax.com/ai-playbook/models/gpt-5-6-luna) — OpenAI, text, proprietary. The cost-efficient tier of OpenAI's GPT-5.6 family — optimized for cost-sensitive, high-volume workloads. Keeps the family's 1.05M-token context window at the lowest price of the three tiers.
- [GPT-5.6 Sol](https://cyberax.com/ai-playbook/models/gpt-5-6-sol) — OpenAI, text, proprietary. The frontier tier of OpenAI's GPT-5.6 family — built for complex professional work: deep reasoning, coding, scientific analysis, and long-running agentic tasks. OpenAI's most capable model, positioned alongside (not as a replacement for) GPT-5.5.
- [GPT-5.6 Terra](https://cyberax.com/ai-playbook/models/gpt-5-6-terra) — OpenAI, text, proprietary. The balanced tier of OpenAI's GPT-5.6 family — the model OpenAI positions as balancing intelligence and cost. Same context window and knowledge cutoff as Sol at half the price.
- [Claude Fable 5](https://cyberax.com/ai-playbook/models/claude-fable-5) — Anthropic, text, proprietary. Anthropic's most capable widely released model — the first public Mythos-class model, a tier above Opus 4.8 for the most demanding reasoning and long-horizon agentic work.
- [Claude Mythos 5](https://cyberax.com/ai-playbook/models/claude-mythos-5) — Anthropic, text, proprietary. Anthropic's restricted frontier model — the same underlying model as Claude Fable 5 with safeguards lifted in select domains. Invitation-only through Project Glasswing, succeeding Claude Mythos Preview.
- [Claude Opus 4.8](https://cyberax.com/ai-playbook/models/claude-opus-4-8) — Anthropic, text, proprietary. Anthropic's most capable Opus-tier model for complex reasoning, long-horizon agentic coding, and high-autonomy work. The flagship of the standard Claude lineup, succeeding Opus 4.7; Claude Fable 5 sits in a new tier above it.
- [Stable Audio 3.0](https://cyberax.com/ai-playbook/models/stable-audio-2-5) — Stability AI, audio-gen, open-weights. Stability AI's open-weight audio model. Generates music tracks, sound effects, and audio loops from text prompts — the open alternative to Suno/Udio. (Small/medium tiers open-weighted; large is API-only.)
- [Gemini 3.5 Flash](https://cyberax.com/ai-playbook/models/gemini-3-5-flash) — Google, text, proprietary. Google's current top Flash-tier Gemini — sustained frontier-level intelligence for agentic and coding tasks at high speed and low cost. The GA successor to the Gemini 3 Flash preview.
- [Gemini 3.1 Flash-Lite](https://cyberax.com/ai-playbook/models/gemini-3-1-flash-lite) — Google, text, proprietary. Google's low-latency Gemini 3-series workhorse for straightforward multimodal tasks at scale. It is designed for high-frequency agent routing, extraction, translation, and summarization work.
- [Gemini 3.1 Flash Live Preview](https://cyberax.com/ai-playbook/models/gemini-3-1-flash-live-preview) — Google, realtime-voice, proprietary. Google's current low-latency audio-to-audio Live API model for real-time dialogue and voice-first applications. It replaces the earlier Gemini Live surface with the Gemini 3.1 stack.
- [Gemini 3.1 Pro Preview](https://cyberax.com/ai-playbook/models/gemini-3-1-pro-preview) — Google, text, proprietary. Google's top Gemini 3-series model for advanced reasoning, coding, and agentic workflows. It improves the Gemini 3 Pro line with better thinking, tool use, and factual consistency.
- [Veo 3.1](https://cyberax.com/ai-playbook/models/veo-3-1) — Google, video-gen, proprietary. Google's flagship video generation model. Adds advanced creative controls and improved prompt adherence on top of the Veo 3 native-audio foundation.
- [Veo 3.1 Lite](https://cyberax.com/ai-playbook/models/veo-3-1-lite) — Google, video-gen, proprietary. Google's efficient, developer-first variant of Veo 3.1. Lower cost and faster generation; same family as the main 3.1 model with reduced fidelity for tighter feedback loops.
- [DeepSeek V4 Flash](https://cyberax.com/ai-playbook/models/deepseek-v4-flash) — DeepSeek, text, open-weights. DeepSeek's efficient tier of the V4 generation. Faster and cheaper than V4 Pro; the practical default for high-throughput agentic workloads.
- [DeepSeek V4 Pro](https://cyberax.com/ai-playbook/models/deepseek-v4-pro) — DeepSeek, text, open-weights. DeepSeek's flagship general-purpose MoE model. Successor to V3; competitive with closed frontier-tier models at open-weights cost.
- [Midjourney V8.1](https://cyberax.com/ai-playbook/models/midjourney-v6) — Midjourney, image-gen, proprietary. Midjourney's flagship image generator. V8.1 is its fastest model yet (2K HD by default); strong artistic quality and a distinctive aesthetic, bound to its Discord and web product, no public API.
- [GPT-5.5](https://cyberax.com/ai-playbook/models/gpt-5-5) — OpenAI, text, proprietary. OpenAI's flagship model — a unified general-purpose and reasoning model positioned as a new class of intelligence for coding and professional work. The current successor to the GPT-4o / o-series lines.
- [GPT Image 2](https://cyberax.com/ai-playbook/models/gpt-image-2) — OpenAI, image-gen, proprietary. OpenAI's current image generation and editing model. It replaces the older DALL·E line with a single state-of-the-art model for image creation and edits.
- [Qwen3.6 35B-A3B](https://cyberax.com/ai-playbook/models/qwen-3-5-122b) — Alibaba, text, open-weights. Alibaba's current open-weight flagship from the Qwen3.6 generation. MoE architecture with 35B total / 3B active parameters; natively multimodal (text + vision) with up to ~1M-token context.
- [Gemma 4 31B](https://cyberax.com/ai-playbook/models/gemma-4-31b) — Google, text, open-weights. Google's flagship open-weight Gemma 4. Natively multimodal (text + vision); supersedes Gemma 2 with significantly improved capabilities and a larger size.
- [Mistral Medium 3.5](https://cyberax.com/ai-playbook/models/mistral-medium-3-5-26-04) — Mistral AI, text, open-weights. Mistral's frontier-class multimodal model for agentic and coding use cases. It sits between the largest flagship tier and the lighter Small line while remaining open-weight.
- [Suno v5.5](https://cyberax.com/ai-playbook/models/suno-v4) — Suno, audio-gen, proprietary. Suno's flagship music generation model. v5.5 adds Voices, Custom Models, and personalization; produces full songs (vocals + instrumentation) from a text prompt — the most polished consumer music AI.
- [GPT-5.4 mini](https://cyberax.com/ai-playbook/models/gpt-5-4-mini) — OpenAI, text, proprietary. OpenAI's strongest small model for coding, computer use, and subagents. The efficient, lower-cost tier of the GPT-5 line and the named replacement for o4-mini.
- [Mistral Small 4](https://cyberax.com/ai-playbook/models/mistral-small-4-0-26-03) — Mistral AI, text, open-weights. Mistral's hybrid open model that unifies instruct, reasoning, and coding in a single efficient line. It is the current small generalist flagship in Mistral's open lineup.
- [Claude 4.6 Sonnet](https://cyberax.com/ai-playbook/models/claude-4-6-sonnet) — Anthropic, text, proprietary. Anthropic's mid-tier model — the practical default for production workloads. Balances quality and cost for most applications.
- [FLUX.2 [dev]](https://cyberax.com/ai-playbook/models/flux-2-dev) — Black Forest Labs, image-gen, open-weights. Black Forest Labs' next-generation open-weight image model. Supersedes FLUX.1 [dev] with improved quality and control; remains the open-weights choice for self-hosted image generation.
- [GPT-5.3-Codex](https://cyberax.com/ai-playbook/models/gpt-5-3-codex) — OpenAI, code, proprietary. OpenAI's most capable agentic coding model — tuned for long-horizon software engineering, tool use, and the Codex agent surface.
- [Kling 3.0](https://cyberax.com/ai-playbook/models/kling-2-0) — Kuaishou, video-gen, proprietary. Kuaishou's video generation model. 3.0 adds native audio, multimodal input, and up to 15-second clips; strong on human motion and physical realism.
- [ElevenLabs Eleven v3](https://cyberax.com/ai-playbook/models/elevenlabs-multilingual-v2) — ElevenLabs, text-to-speech, proprietary. ElevenLabs' flagship expressive TTS (Eleven v3). The benchmark for natural-sounding speech and voice cloning across 70+ languages; Flash/Turbo v2.5 cover low-latency use.
- [Voyage 4](https://cyberax.com/ai-playbook/models/voyage-3) — Voyage AI, embeddings, proprietary. Voyage AI's flagship embedding model (voyage-4-large). Top of MTEB across many tasks; the embedding service Anthropic recommends for Claude RAG workloads.
- [OCR 3](https://cyberax.com/ai-playbook/models/ocr-3-25-12) — Mistral AI, vision, proprietary. Mistral's current OCR service for its Document AI stack. It extracts interleaved text and images from documents and replaces the older Mistral OCR line.
- [Cohere Rerank v4](https://cyberax.com/ai-playbook/models/cohere-rerank-v3) — Cohere, reranking, proprietary. Cohere's flagship reranker (rerank-v4.0-pro). The standard second-pass model after a vector or BM25 retrieval — 100+ languages, 32K per-doc context, bumps precision noticeably with minimal architecture changes.
- [Runway Gen-4.5](https://cyberax.com/ai-playbook/models/runway-gen-3) — Runway, video-gen, proprietary. Runway's flagship video generation model. Strong creative-tooling ecosystem (motion brush, camera control, style transfer); the production tool of choice for many video creators.
- [Devstral 2](https://cyberax.com/ai-playbook/models/devstral-2-25-12) — Mistral AI, code, open-weights. Mistral's frontier code-agents model for software engineering tasks. It is designed for tool-heavy coding workflows across whole repositories and multi-file edits.
- [Ministral 3 14B](https://cyberax.com/ai-playbook/models/ministral-3-14b-25-12) — Mistral AI, text, open-weights. The largest model in Mistral's Ministral 3 family. It is built for local deployment with strong text and vision performance on diverse hardware.
- [Mistral Large 3](https://cyberax.com/ai-playbook/models/mistral-large-3) — Mistral AI, text, proprietary. Mistral AI's flagship model. Successor to Mistral Large 2 with improved multilingual coverage and reasoning. EU-jurisdiction provider.
- [FLUX.2 [pro]](https://cyberax.com/ai-playbook/models/flux-1-pro) — Black Forest Labs, image-gen, proprietary. Black Forest Labs' commercial flagship (FLUX.2 [pro]). The closed top tier when you need commercial-use rights and maximum quality; FLUX.2 [dev] is the open-weight sibling.
- [Claude Haiku 4.5](https://cyberax.com/ai-playbook/models/claude-haiku-4-5) — Anthropic, text, proprietary. Anthropic's fast, cheap tier. The right choice for high-throughput agentic work and tasks where latency matters more than depth.
- [Luma Ray3](https://cyberax.com/ai-playbook/models/luma-dream-machine) — Luma AI, video-gen, proprietary. Luma AI's flagship video model (behind the Dream Machine app). Ray3 is reasoning-driven with native HDR; strong on cinematic camera moves and 3D-aware generation.
- [Codestral 25.08](https://cyberax.com/ai-playbook/models/codestral-25-08) — Mistral AI, code, proprietary. Mistral's current code-completion model, released at the end of July 2025. It is tuned for low-latency fill-in-the-middle and high-frequency code generation tasks.
- [Llama 4 Maverick](https://cyberax.com/ai-playbook/models/llama-4-maverick) — Meta, text, open-weights. Meta's flagship Llama 4 model — natively multimodal, larger MoE architecture than Scout. The Llama 4 frontier-tier entry.
- [Llama 4 Scout](https://cyberax.com/ai-playbook/models/llama-4-scout) — Meta, text, open-weights. Meta's small/efficient Llama 4 variant — natively multimodal MoE architecture. The practical Llama 4 entry point for self-hosted multimodal applications.
- [NVIDIA Parakeet](https://cyberax.com/ai-playbook/models/nvidia-parakeet) — NVIDIA, speech-to-text, open-weights. NVIDIA's STT family. parakeet-tdt-0.6b-v2 tops the HF Open ASR leaderboard for English; parakeet-tdt-0.6b-v3 adds 25-language multilingual support. Very fast on NVIDIA hardware via NeMo.
- [Cohere Embed v4](https://cyberax.com/ai-playbook/models/cohere-embed-v3) — Cohere, embeddings, proprietary. Cohere's flagship embedding model (embed-v4.0). Multimodal (text + image) with up to 128K context; strong multilingual coverage and compressed (int8/binary) embeddings — useful for cost-sensitive RAG.
- [GPT-4.1](https://cyberax.com/ai-playbook/models/gpt-4-1) — OpenAI, text, proprietary. OpenAI's developer-first frontier model for coding, instruction following, and long-context work. It is the API-oriented successor line to older GPT-4 variants.
- [GPT-4.1 mini](https://cyberax.com/ai-playbook/models/gpt-4-1-mini) — OpenAI, text, proprietary. OpenAI's smaller GPT-4.1 variant. It keeps the 1M-token context window while lowering cost and latency enough for high-volume agent and application workloads.
- [Gemini 2.5 Flash](https://cyberax.com/ai-playbook/models/gemini-2-5-flash) — Google, text, proprietary. Google's speed-optimised tier. Cheap and fast multimodal, with a generous free tier on AI Studio for prototyping.
- [Gemini 2.5 Pro](https://cyberax.com/ai-playbook/models/gemini-2-5-pro) — Google, text, proprietary. Google's flagship multimodal model. Massive context window and competitive frontier-tier performance, with extended thinking on demand.
- [GPT-4o mini TTS](https://cyberax.com/ai-playbook/models/gpt-4o-mini-tts) — OpenAI, text-to-speech, proprietary. OpenAI's current text-to-speech model, built on GPT-4o mini. It replaces the older tts-1 line with better quality and a newer multimodal stack.
- [GPT-4o Transcribe](https://cyberax.com/ai-playbook/models/gpt-4o-transcribe) — OpenAI, speech-to-text, proprietary. OpenAI's hosted speech-to-text model, built on GPT-4o. The API-recommended transcription model, with lower word-error rate and better language recognition than the original Whisper API.
- [Udio Allegro v1.5](https://cyberax.com/ai-playbook/models/udio) — Uncharted Labs, audio-gen, proprietary. Suno's main competitor. Allegro v1.5 is the current model — faster generation with strong genre control and a focus on song-structure quality.
- [Pika 2.2](https://cyberax.com/ai-playbook/models/pika-2) — Pika Labs, video-gen, proprietary. Pika's video generation model. 2.2 adds Pikaframes keyframe transitions and 1080p; differentiates with Scene Ingredients (drop in characters/objects across shots) for character consistency across clips.
- [DeepSeek R1](https://cyberax.com/ai-playbook/models/deepseek-r1) — DeepSeek, reasoning, open. DeepSeek's open-weight reasoning model. Released with full weights and a permissive MIT license — the first competitive open reasoning model.
- [Phi-4](https://cyberax.com/ai-playbook/models/phi-4) — Microsoft, text, open. Microsoft's small model trained heavily on synthetic data. Punches above its 14B weight on reasoning and math; MIT-licensed and runs locally.
- [Llama 3.3 70B](https://cyberax.com/ai-playbook/models/llama-3-3-70b) — Meta, text, open-weights. Meta's latest 70B-parameter open-weight model. Reaches frontier-tier performance for English-centric tasks while remaining self-hostable.
- [Qwen2.5-Coder 32B](https://cyberax.com/ai-playbook/models/qwen-2-5-coder-32b) — Alibaba, code, open. Alibaba's flagship open code model. 32B parameters and Apache 2.0 — the strongest open coding model that fits on a single workstation GPU.
- [Stable Diffusion 3.5](https://cyberax.com/ai-playbook/models/stable-diffusion-3-5) — Stability AI, image-gen, open-weights. Stability AI's open image-gen family. Three sizes (Large, Large Turbo, Medium) — runs locally on consumer GPUs and supports a massive ecosystem of LoRAs and ControlNets.
- [Whisper large-v3 Turbo](https://cyberax.com/ai-playbook/models/whisper-large-v3-turbo) — OpenAI, speech-to-text, open. OpenAI's distilled Whisper variant. ~8× faster than large-v3 with most of the accuracy retained — the practical default for high-throughput STT pipelines.
- [Llama 3.2 Vision](https://cyberax.com/ai-playbook/models/llama-3-2-vision) — Meta, vision, open-weights. Meta's open-weight vision-language family. 11B and 90B variants — the practical open-weights vision model for self-hosted multimodal applications.
- [Qwen 2.5 72B](https://cyberax.com/ai-playbook/models/qwen-2-5-72b) — Alibaba, text, open-weights. Alibaba's flagship open-weight model. Strong on coding, math, and Chinese-language tasks; competitive with Llama 3.3 70B on Western benchmarks.
- [Qwen 2.5 7B](https://cyberax.com/ai-playbook/models/qwen-2-5-7b) — Alibaba, text, open. Alibaba's small-tier open model. Apache-licensed, runs on consumer hardware, and remains competitive with other 7B-class models on coding and math.
- [Llama 3.1 8B](https://cyberax.com/ai-playbook/models/llama-3-1-8b) — Meta, text, open-weights. Meta's small open-weight model. Runs on consumer hardware (16GB GPU or modern laptop) and remains a strong default for local-first AI.
- [GPT-4o mini](https://cyberax.com/ai-playbook/models/gpt-4o-mini) — OpenAI, text, proprietary. OpenAI's small, cheap, fast frontier model. The default workhorse for high-volume tasks where GPT-4o would be overkill.
- [DeepSeek-Coder V2](https://cyberax.com/ai-playbook/models/deepseek-coder-v2) — DeepSeek, code, open-weights. DeepSeek's open code model. MoE architecture with strong coverage across 338 programming languages; the open-weights coder of choice for high-end self-hosting.
- [BGE Reranker v2](https://cyberax.com/ai-playbook/models/bge-reranker-v2) — BAAI, reranking, open. BAAI's open reranker. Apache 2.0 weights, multiple sizes, and the open-weights default for self-hosted RAG pipelines that need a second-pass.
- [StarCoder2 15B](https://cyberax.com/ai-playbook/models/starcoder2-15b) — BigCode, code, open-weights. BigCode's collaborative code model. 15B parameters trained on 600+ programming languages; strong fit for IDE completion and self-hosted code search.
- [BGE-M3](https://cyberax.com/ai-playbook/models/bge-m3) — BAAI, embeddings, open. BAAI's multi-functional embedding model. Supports dense, sparse, and multi-vector retrieval in one model — the strongest open-weights embedding option.
- [OpenAI text-embedding-3-large](https://cyberax.com/ai-playbook/models/openai-text-embedding-3-large) — OpenAI, embeddings, proprietary. OpenAI's largest embedding model. 3072 dimensions, multilingual, and the default high-quality option for RAG and semantic search.
- [OpenAI text-embedding-3-small](https://cyberax.com/ai-playbook/models/openai-text-embedding-3-small) — OpenAI, embeddings, proprietary. OpenAI's small embedding tier. 1536 dimensions; the cheap default for most RAG and semantic-search workloads where quality is sufficient.
- [Magnific AI](https://cyberax.com/ai-playbook/models/magnific-ai) — Magnific, image-enhance, proprietary. The premium 'creative upscaler'. Invents detail at high zoom factors using diffusion priors — the right choice when you want an upscale that adds fidelity rather than just enlarging pixels.
- [Whisper large-v3](https://cyberax.com/ai-playbook/models/whisper-large-v3) — OpenAI, speech-to-text, open. High-accuracy multilingual speech-to-text. Best-in-class for non-English audio; the de-facto open baseline.
- [Coqui XTTS v2](https://cyberax.com/ai-playbook/models/coqui-xtts-v2) — Coqui, text-to-speech, open-weights. Coqui's open multilingual TTS. Supports voice cloning with a 6-second sample across 17 languages — the leading open alternative to ElevenLabs.
- [Distil-Whisper](https://cyberax.com/ai-playbook/models/distil-whisper) — Hugging Face, speech-to-text, open. Hugging Face's distillation of Whisper (distil-large-v3.5). Faster than Whisper-large-v3-Turbo on long-form audio at small accuracy cost; English-only — pick when you don't need multilingual.
- [Piper](https://cyberax.com/ai-playbook/models/piper) — Rhasspy, text-to-speech, open. A fast, local TTS designed for Raspberry Pi-class hardware. Powers most self-hosted voice assistants where Whisper handles input and Piper handles output.
- [Stable Diffusion x4 Upscaler](https://cyberax.com/ai-playbook/models/stable-diffusion-x4-upscaler) — Stability AI, image-enhance, open-weights. Stability AI's diffusion-based 4× upscaler. Trades speed for quality — invents plausible high-frequency detail rather than just sharpening, which suits AI-generated images especially well.
- [SwinIR](https://cyberax.com/ai-playbook/models/swinir) — ETH Zürich, image-enhance, open. Transformer-based image restoration model. Strong on text, edges, and faces — often produces sharper results than GAN-based upscalers on photographic content.
- [Real-ESRGAN](https://cyberax.com/ai-playbook/models/real-esrgan) — Tencent ARC Lab, image-enhance, open. The de-facto open-weights image upscaler. Battle-tested across years of community use; runs on CPU or GPU, integrates with virtually every local image pipeline.
- [Vosk](https://cyberax.com/ai-playbook/models/vosk) — Alpha Cephei, speech-to-text, open. An offline, lightweight speech recognition toolkit. Runs on phones, Raspberry Pi, and embedded devices — the right choice when Whisper is too heavy.
- [Topaz Gigapixel AI](https://cyberax.com/ai-playbook/models/topaz-gigapixel-ai) — Topaz Labs, image-enhance, proprietary. Industry-standard desktop application for photo upscaling and restoration. The default tool for archival work, photo restoration, and print-sized enlargements where fidelity to the original matters.