Tools.

Cowork is Claude's computer-use workspace — the third tab in the Claude desktop app. Briefing folder, filesystem access, scheduled tasks. Dispatch lets you control it from your phone.

• +The briefing folder (about-me, writing style, do-not-do) makes your style reproducible.
• +Computer use + persistent workspace + mobile remote control in one product.
• –Computer use is still a research preview; multi-app tasks succeed about 50 % of the time in practice.

US parent, EU routing from Sonnet 4.6 on (+10 % multiplier)

Part of the inzpyre starter stack

OpenAI's computer-use product — drives the browser autonomously, researches, clicks, fills forms, plans tasks. Included in the Plus and Pro plans, with per-tier quotas.

• +Deep integration with Custom GPTs, Connectors, Projects.
• +Solid for structured research and booking tasks.
• –Sandboxed browser — no access to local desktop apps.
• –Per-plan quotas, no multi-user setup under Plus.

US parent, Azure routing available for EU

Perplexity's AI browser with computer use. Cited-source research, deep research, computer-use actions right in the tab. Multi-model choice (Claude, GPT, Gemini). Available with Perplexity Pro.

• +AI browser with source tracking — research stays verifiable.
• +Multi-model choice per query.
• –Requires its own browser — if you live in your main browser, you'll keep two windows open.
• –Weaker at writing with style than Claude or ChatGPT.

US company

Voice dictation in any text field (mail, Slack, Notion, IDE). Strips filler words automatically and adapts the tone to the target app. Command mode for voice edits.

• +4x faster than typing, the only major option across macOS, Windows, iOS, and Android.
• –Accuracy drops to 88 % in noisy settings. For pure dictation, $15 feels steep.

US company, audio processed in the cloud

A CLI coding agent with codebase indexing, MCP support, sub-agents. The strongest all-rounder for production cleanup after vibe-code and for serious engineering work.

• +Full control, skills + plugins, runs locally with real filesystem access.
• –No visual builder; you have to be able to read code.

US parent, local execution

Part of the inzpyre starter stack

Conversational full-stack builder with native Supabase integration. You describe the app, Lovable builds the frontend, auth, DB, and deploys.

• +Cleanest UI code among the full-stack builders, Lovable Cloud = managed Supabase.
• –Gets you to 70 %; the last 30 % cost nerves. RLS pitfalls, auth flows break quickly.

Supabase Cloud, EU region selectable

Part of the inzpyre starter stack

In-browser builder from StackBlitz, runs on WebContainer. Drop in a prompt and you get a full web app live in the browser with no setup.

• +The fastest „idea to live app“ path; 30 minutes for a POC is realistic.
• –Token use scales heavily with project size, and code quality is below v0 and Lovable.

US company, Supabase EU works as a backend

UI-focused code generator. Produces idiomatic React/Next.js code with Tailwind and shadcn/ui. Since February 2026 with a Next.js sandbox, Git integration, and DB connectors.

• +Best UI code quality on the market, native shadcn conventions, clean Git workflow.
• –Heavily frontend-focused; backend and auth feel less first-class. Token pricing has been hard to read since 2026.

Vercel hosting, EU functions selectable

Part of the inzpyre starter stack

Cloud IDE plus agent mode. Agent 3 (2026) builds whole apps including hosting, DB, and deploy in one environment.

• +Lowest barrier to entry; everything in one place.
• –Lock-in on export, Replit DB is a footgun without an active subscription, pricing complaints around checkpoint credits.

US hosting

AI-first code editor (a VS Code fork). Composer for multi-file edits, deep codebase indexing, agent mode with background tasks.

• +The best editor for real engineering refactors and large codebases.
• –Trustpilot 1.7/5 because of pricing changes and unclear overages. Not for non-techies.

US company

Mobile app builder via prompt. Builds React Native apps with Expo, ships iOS and Android from one prompt, with full code export and EAS submit integration.

• +The most accessible path from idea to a real store submission. The code is yours.
• –Native modules outside Expo are tricky; credit loops on bugs get expensive.

US company

Postgres, auth, storage, realtime, edge functions in one dashboard. Open source, self-host possible. The default pick for AI builders (Lovable, Bolt, v0).

• +A real SQL DB, EU region Frankfurt available, huge community, pgvector built in.
• –US parent (residual CLOUD-Act risk), 100k MAU cap on Pro, edge-function cold starts.

EU region Frankfurt, Dublin, Paris, Stockholm. DPA available.

Part of the inzpyre starter stack

Google's BaaS — Firestore (NoSQL), Auth, Cloud Functions, hosting, FCM push, Crashlytics. Tightly integrated with Vertex AI and Gemini.

• +Best mobile SDKs in the business, realtime sync „just works“.
• –Firestore NoSQL — complex joins hurt. „Firebase nightmare bills“ on bad queries. Vendor lock-in runs high.

Frankfurt + Belgium selectable, US parent

Reactive backend „as code“ — schema, queries, mutations, and cron jobs all in TypeScript. Live sync without doing WebSockets by hand.

• +Best DX if you already speak TypeScript, optimistic updates and realtime built in.
• –Proprietary data model, no Postgres, high lock-in. EU hosting story is weak (US AWS is the default).

Primarily US AWS, no EU default

Serverless Postgres with Git-style branching, scale-to-zero, compute separated from storage. Part of Databricks since May 2025.

• +Branching for a DB-per-PR is brilliant, scale-to-zero saves money on hobby projects.
• –DB only (no auth/storage), 1–3 s cold start on wake-up.

EU eu-central-1 Frankfurt, Databricks (US) as parent

Managed NoSQL database (document store). Atlas Vector Search integrated since 2024 for RAG workloads.

• +Mature vector search, many language SDKs, solid for document-driven data.
• –The document model fits relational domains badly. More complex than Supabase for beginners.

Frankfurt, Dublin, and many more on AWS/Azure/GCP

The native home of Next.js. Edge network, ISR, image optimisation, preview deployments, AI SDK + AI Gateway built in.

• +Best DX for Next.js, preview URLs per branch, the default deploy target for Lovable/Bolt/v0.
• –Notorious surprise bills on viral posts. Bandwidth overage costs more than Cloudflare. The Hobby plan is forbidden for commercial use.

Functions pinnable to fra1, dub1, cdg1, arn1

Part of the inzpyre starter stack

Jamstack-hosting pioneers. Static + functions + forms + identity. Since April 2026 credit-based with unlimited team members.

• +Unlimited team members is a killer feature for agencies, solid build pipeline.
• –Hard to forecast credits, Next.js support not as smooth as Vercel.

Global CDN, some functions on EU edge

Static hosting with unlimited bandwidth, Workers for dynamic logic, Pages Functions. Combined with R2 + D1 + KV you get a complete stack.

• +No bandwidth limit on any tier; going viral is free. Best EU data-localisation suite.
• –Next.js doesn't run 1:1 (no full RSC support without workarounds), DX is less smooth than Vercel.

Edge across Europe, Regional Services add-on for EU-only

Part of the inzpyre starter stack

Heroku replacement. Deploys containers/apps straight from GitHub. Postgres, Redis, MySQL built in. Per-second usage billing.

• +Best DX for full-stack apps with a backend server (Express, FastAPI, Rails).
• –Costs scale fast on always-on services. Smaller vendor, less compliance depth.

EU europe-west4 (Netherlands)

Global container platform, deploys apps in 30+ regions. Firecracker VMs, real multi-region deploys.

• +Real multi-region deploys easier than anywhere else.
• –No free tier, trial is tiny (2 VM-h). Steep learning curve (fly.toml, fly CLI).

Frankfurt, Amsterdam, Paris, Stockholm, Madrid, Warsaw

Opus 4.7 / Sonnet 4.6 / Haiku 4.5 — reasoning, code, agents, tool use very robust. 1M context standard from 4.6 on.

• +Prompt caching gives a 90 % discount, tool use state of the art, Batch API -50 %.
• –The Opus 4.7 tokenizer needs up to 35 % more tokens for the same text vs. 4.6. EU routing adds a 10 % multiplier.

EU routing optional, AWS Bedrock and GCP Vertex available

GPT-5.5 / 5.4 / Mini / Nano — top intelligence, native computer use, 1M context, function calling very mature.

• +Largest ecosystem, native computer use, cached input -90 % automatically.
• –GPT-5.5 Pro at $30/$180 is noticeably pricier. Data residency only via Azure/Foundry.

Azure routing for EU

Gemini 3.5 Flash / 3.1 Pro Preview / 2.5 Flash-Lite — strong multimodal, search grounding, up to 2M context.

• +Best multimodal input (image, video, audio native), search grounding built in.
• –Pricing jumps above 200k context. Nano Banana 2 Personal Intelligence is a GDPR trigger.

Vertex AI regional endpoints, Frankfurt available

Grok 4.3 / 4.20-reasoning / Multi-Agent — an all-rounder with native access to real-time X/Twitter data.

• +Cheap in the top tier, native X integration for real-time data.
• –Smaller ecosystem, model names change often.

US company

V4-Flash / V4-Pro — a very cheap reasoning specialist with explicit JSON mode, thinking mode, and FIM.

• +Unbeatable price for structured tasks, cache hit at one tenth of launch price.
• –Often weaker at German prose. Chinese vendor — handle sensitive data with care.

Hosting in China

An umbrella for every Google service API. Common in builds: Maps and Places (POI data, geocoding), Drive (file sync), Calendar (events), YouTube Data (channel and video metadata), Translate (translations), Vision (OCR, image analysis). Auth flows through the Google Cloud Console (API key or OAuth).

• +One auth system across many services. Very stable, long-running APIs with strong SDK coverage.
• –Quotas and billing differ per API; worst case you watch several cost centres. GDPR setup isn't trivial for Maps/Analytics.

US parent, regional endpoints configurable (e.g. europe-west)

Marketplace with thousands of third-party APIs under one auth. Examples: LinkedIn profile API (used in the AI Disruption Score), Instagram, Twitter/X, web scraping, sports data, weather, translation. One account, one API key, instant access to the whole marketplace.

• +Very fast to get started — one auth layer for many data sources. Handy when the official API is expensive or fiddly.
• –Quality varies per provider; some endpoints are grey-zone (e.g. social-media scraping against ToS). Latencies often worse than direct APIs.

US company

Stock images and stock videos, free with attribution. One API key and you're going. Good for MVP images, hero sections, marketing material, and demo content.

• +Completely free for most use cases. Good quality, curated content, clear licence.
• –The limit can run out on viral apps. Selection is smaller than premium libraries (Getty, Shutterstock).

US company

Premium image generator with the most aesthetic default look on the market. V8.1 (April 2026) is 4–5x faster than V7. Style codes, moodboards, raw mode.

• +Aesthetics out of the box are hard to beat, native 2K.
• –No open self-service API, text rendering trails Ideogram/GPT Image 2. Companies above $1M revenue must use Pro/Mega.

US company

A family of models. Pro = the 32B flagship via API, [klein] 4B is Apache-2.0 and self-hostable.

• +Best price-to-quality ratio over API, very strong at photorealism, self-hosting is possible.
• –No first-party end-user UI; you'll need fal, Replicate, Krea, or your own wrapper.

Black Forest Labs is a German startup (Freiburg); hosting depends on the provider you pick

Part of the inzpyre starter stack

Google's new default image generator (February 2026). First-class text rendering (#2 behind GPT Image 2), world knowledge, Personal Intelligence mode.

• +Free and strong, fast iteration workflow, pulls fresh web info into renders.
• –Personal Intelligence is a GDPR trigger for photo uploads. Aesthetics sometimes feel more generic than Midjourney.

Google, US parent

Part of the inzpyre starter stack

The DALL-E successor (April 2026). Native reasoning inside the image model, #1 on the Image Arena at 1512 ELO. 99 % text accuracy in English.

• +Best text rendering in the world, up to 8 coherent images with character consistency, 2K resolution.
• –Premium pricing via API, sometimes too „polished“ for an artist look.

US parent

Specialist for text-in-image and graphic design. 90–95 % text accuracy. Style references (up to 3 images), canvas editor.

• +The best option for posters, flyers, and social posts with text. Four style types, three speed tiers.
• –Photorealism is solid, but not at FLUX/Nano-Banana level.

US company

Adobe's generative AI stack. Trained only on Adobe Stock and public domain. The only provider with IP indemnification (Enterprise).

• +Licence-clean training data, automatic C2PA Content Credentials, Photoshop integration.
• –Image quality lags Midjourney/FLUX/GPT Image 2. Reads more „safe“ than „wow“.

US parent, Enterprise with EU data residency available

The industry standard for AI video. Current Gen-4 models for text-to-video, image-to-video, video inpainting, and style transfer. Plus editing tools (motion brush, camera control). Also available as an API.

• +Best all-rounder with an editing toolkit. Very production-ready UI, lots of pro features (camera control, motion brush).
• –Credits model: heavy generations burn through them fast. The API costs more than a subscription.

US company

Google's video model with high motion quality and longer clips (as of May 2026: the current generation). Available in the Gemini app, in Vertex AI, and via the Gemini API.

• +Very strong at coherent motion and realistic physics. Deeply integrated into the Google ecosystem.
• –Generation takes relatively long. The per-plan credit system is hard to read.

Vertex AI regional endpoints, EU region selectable

OpenAI's video model. Included in ChatGPT Plus/Pro, plus API access. Strong on photorealistic scenes, with a storyboarding workflow right in the web UI.

• +Very aesthetic default output, good for ad and story clips. Storyboard editor right in the UI.
• –Quotas on Plus/Pro are tight. Long clips remain expensive. Regional rollout is uneven.

US parent, Azure routing available for EU

A fast, cheap video generator focused on social clips and effects (Pika Effects, lip-sync, scene ingredients). Web UI plus a Discord workflow.

• +Lowest entry price in the top tier. The effects library for social content is playful.
• –Motion quality below Runway/Veo, resolution capped on the Standard plan.

US company

Cinematic clips with especially strong camera moves (tracking, orbit, push). Web UI, fast generation, low barrier to entry. Also available as an API.

• +Best camera moves out of the box, cinematic look without heavy prompt engineering.
• –Character consistency across clips is still weak. No built-in editing toolkit.

US company

A Chinese video model with very strong motion consistency and realistic physics. Longer clips than most Western providers, high-resolution output.

• +Motion and character consistency often lead the field. Long clips (up to 10 seconds+) in the standard tier.
• –China hosting: be careful with person- or business-sensitive content. UI partly Chinese-only.

Hosting in China (Kuaishou)

Apple's official IDE for iOS, iPadOS, macOS, watchOS, tvOS, and visionOS. The required toolchain for App Store submission of Swift apps.

• +The only official toolchain for native Apple platforms. Simulator, Instruments, and TestFlight integrated.
• –macOS only, slow builds, forced updates with new iOS releases.

Apple, US parent, builds local on your machine

React Native framework with OTA updates, config plugins, and native modules. The default stack for cross-platform apps in 2026.

• +Fastest cross-platform workflow (iOS + Android), live reload via Expo Go, OTA updates without store review.
• –Native modules outside the Expo universe need custom builds. Performance trails native Swift/Kotlin.

US company, EAS builds in the cloud

Build and submit pipeline for Expo apps. Produces iOS/Android binaries in the cloud and uploads them straight to App Store and Play Store.

• +Saves a local Xcode setup for plain iOS submission, OTA updates save review wait times.
• –Cloud builds may queue on the free tier; EAS submit still requires Apple and Google accounts.

US cloud

Open-source workflow engine with a visual node editor and native AI-agent nodes. Make/Zapier you can self-host.

• +Predictable costs (especially self-hosted), no task counter, native LangChain concepts, huge community.
• –Steep learning curve; self-hosting means responsibility (updates, backups, HTTPS).

Self-host on Hetzner possible (EU), cloud region selectable

Part of the inzpyre starter stack

Visual scenario builder, sitting between Zapier-simple and the deeper n8n end. Very strong mapping and iterator concept.

• +Best price-to-performance ratio for classic SaaS integration. Bring-your-own-key for LLMs since November 2025.
• –Ops counter is tricky — a CRM↔Sheet sync can eat 50 ops per run. No self-host.

Czech company, EU hosting

„When trigger A, then action B“ taken to the simplest form. Largest app directory with 8,000+ integrations.

• +Fastest aha moments, very clean docs, templates for almost every standard workflow.
• –Expensive at scale. Anyone running >1,000 tasks/month should price out Make or n8n — quickly 3–5x cheaper.

US company

A background-jobs platform for TypeScript. Code lives in your repo, version controlled, with resumable runs and built-in caching. The best option for long-running agents (>30 min).

• +Code lives in the repo (not in someone else's UI), first-class for long agent loops.
• –Requires dev skill; no visual builder.

US company

Code-first automation with a visual wrapper. Node.js/Python snippets between app steps, with per-step inspect and replay.

• +Best bridge between no-code and code, true inspect/replay, generous AI token volume on the free tier.
• –Without JS or Python knowledge you'll hit a wall on mid-complexity. The daily credit reset can hurt.

US company

An AI-native agent platform. You build an agent with tools (mail, calendar, phone, web browse) and give it a job description; the workflow falls out of the agent loop.

• +Real agent loops with memory and self-correction. A „sales assistant that books meetings“ is a one-hour setup.
• –Black-box behaviour on hallucinations, voice calling burns credits fast, less depth on classic SaaS integrations than n8n/Make.

US company

Meeting-notes app that runs locally on your Mac — no bot in the participant list, very clean notes, and automatic action items.

• +No bot in the call list, local audio processing.
• –Mac only (as of May 2026), free plan heavily limited.

US company, local processing

AI-driven deck and doc builder. From one idea or prompt, Gamma builds a presentation-ready deck with images, layouts, and animations in seconds.

• +Fastest idea-to-deck pipeline on the market.
• +Live-editable output, no PowerPoint export ping-pong.
• –Output style often looks generic and recognisable without customisation.

US company

A keyboard-first launcher for macOS with built-in AI chat, AI commands, and Quick AI on a hotkey. A Spotlight replacement for Macs.

• +Hotkey speed, extension ecosystem, cloud sync across Macs.
• +Model choice in one inbox (GPT/Claude/Gemini).
• –macOS-first (Windows in beta).
• –Advanced AI feels redundant for most.

US company