The 135 best AI Platforms tools

RANK

Compare CE, EE and Cloud using your required features, production license, relevance checks and total operating cost.

Pros: Search, filtering and hybrid retrieval through one API · A separately identified MIT Community Edition

RANK

Select a supported model/hardware stack, then validate its API behavior and operating budget before production.

Pros: Documented API and model-serving controls · Open engine code with configurable deployment

RANK

Choose Framer after checking hosting control, editing/AI budget and the workspace's applicable training settings.

Pros: Design, CMS and publishing in one documented workflow · A documented proxy option for delivery integration

RANK

Check the exact model's rights and backend support, then measure memory, quality and operating cost for your deployment.

Pros: Source-level control and documented CLI/server paths · Multiple backend and quantization choices

RANK

Choose the local or cloud path, verify App/model rights and resource limits, then assess the workflow with your own data.

Pros: Local model and documented local-server workflows · Free local use with optional cloud choices

RANK

Check organization eligibility, content ownership/hosting and the combined answer/update budget before selecting a plan.

Pros: Documentation sources and publishing workflow can stay connected · Separate controls for AI credit limits and overages

RANK

Keep local execution, model permissions and cloud billing separate, then assess one defined task with the chosen configuration.

Pros: Local execution with a documented local-only setting · Choice between local compute and hosted model plans

RANK

Choose it when you can operate the deployment and respect release-specific branding and model-provider conditions.

Pros: Self-hosted interface with several model connections · RAG and user-management documentation

RANK

Choose a deployment after evaluating your data, embedding rights, recall requirements and complete infrastructure/cloud bill.

Pros: Vector retrieval with metadata conditions · Multiple deployment options

RANK

Evaluate search quality and resource needs, then choose GPL-compliant operation or the appropriate Cloud contract.

Pros: Several search modes and scoped keys · Server and managed deployment choices

RANK

A candidate for editable application workflows; select the edition, check credit rules and map every external data destination before deployment.

Pros: Documented branches, loops and code steps · MIT core can be self-hosted

RANK

Consider it if your team can maintain queries and operations; confirm edition rights, outbound calls and full-month user billing.

Pros: Self-hostable Apache-2.0 CE · UI, queries and code in one builder

RANK

Consider it with an owner for examples and scorers; check cost meters, architecture and contract before relying on evaluation scores.

Pros: Documented experiment/scorer workflow · Enterprise data-plane deployment choice

RANK

Consider Browserbase when you can maintain automation logic, authorize the target workflow and govern saved session data. Estimate browser time and each metered extra per task before choosing a plan.

Pros: Documented integration paths for existing automation clients · Live interaction and replay for inspecting sessions

RANK

Choose the deployment and embedding provider alongside the database.

Pros: Several retrieval methods · Apache-licensed self-hosting option

RANK

Quote the actual workload and contract; component openness is not a platform license.

Pros: Data and model work share a platform · Documented catalog governance

RANK

Check the edition's license and every connected data processor before publishing.

Pros: Visual orchestration with retrieval · Several deployment routes

RANK

Compare the exact endpoint, billing unit and supplemental agreement.

Pros: Several API calling patterns · Specific compute and API price units

RANK

Evaluate the exact Glide product, data source and access rules before budgeting a business app.

Pros: Documented table, component and workflow structure · Row-level access controls and plan-specific connectors

RANK

Budget the host and models, and secure the functions you expose.

Pros: Function-based UI building · Local and hosted deployment choices

RANK

Consider Groq for a model-specific API integration after checking usage costs, model terms and retention settings.

Pros: Documented compatible API and model catalog · Separate usage billing and budget controls

RANK

Check the chosen model, hardware and every enabled external connection.

Pros: Local-model deployment option · Cloud and tool connections are configurable

RANK

Compare core, Enterprise and Cloud rights alongside trace retention and access.

Pros: Tracing and evaluation in one workflow · Core can be self-hosted

RANK

Choose the operator and model contracts alongside the chat UI.

Pros: Model endpoints can be configured · Multi-user and conversation workflows

RANK

Budget Enterprise capacity separately from provider inference and review every callback.

Pros: Centralized keys and usage controls · SDK and gateway deployment choices

RANK

Choose the deployment first and price infrastructure and model dependencies separately.

Pros: Several deployment modes · Filtering and multiple retrieval methods

RANK

Evaluate Modal for hosted programmable compute after separating compute seconds, warm time and token endpoints.

Pros: Execution, resource configuration and deployment in code · Documented per-resource prices and retention scopes

RANK

Consider pgvector when PostgreSQL is a suitable operational base; validate indexing and costs rather than assuming the cheapest or most scalable architecture.

Pros: Vector and relational queries in one PostgreSQL system · Permissive extension licence

RANK

Use PyTorch as a model-development framework after checking hardware, dependencies and the licences of the actual models and data.

Pros: Composable tensors, autograd and model-development modules · Redistributable source and binary code under stated conditions

RANK

Consider Retool when someone can own data access, generated code review and the combined seat/usage budget.

Pros: Documented resource and app-building workflows · Separate builder and internal-user pricing

RANK

Consider Softr when a hosted, configurable portal fits the workflow; price the actual users and AI use, and confirm rights and exit requirements before committing.

Pros: AI and visual editing share app structure · Native login, data and access controls are documented

RANK

Choose browser and model routes deliberately; free SDK code does not include hosted sessions or permission to access every website.

Pros: Mix explicit browser code with model-guided steps · MIT SDK with separately configurable browser and model

RANK

Choose Streamlit for Python-built interfaces, then select hosting whose data and operating terms fit the app.

Pros: Python UI elements for data and chat applications · Library and hosting can be evaluated separately

RANK

Consider WRITER when governed enterprise workflows fit the contract, with seats, tokens and connected services budgeted separately.

Pros: Platform controls across models, data and agents · Published model and feature usage rates

RANK

Consider Algolia when you can maintain indices and model both request and record costs before enabling extra AI services.

Pros: API, indexing and UI integration guides · Configurable relevance and separate usage meters

RANK

Select the deployment arrangement first, then compare the full resource bill and the exact model/data contract.

Pros: Documented function, endpoint and resource configuration · Hosted, BYOC and self-hosted choices

RANK

Evaluate the exact model, region and operating route; Bedrock is a managed platform with several billing and retention regimes.

Pros: AWS-integrated access to different model providers · Documented model-specific billing and data controls

RANK

Evaluate BentoML for packaging and serving, and obtain the exact managed-service architecture, pricing and data terms separately.

Pros: Service code, dependencies and model artifacts packaged together · Apache-2.0 library with container deployment paths

RANK

Choose Bubble when its hosted visual runtime fits the project, and budget workload, AI credits and a possible logic rebuild on migration.

Pros: Interface, database, workflows and hosting in one platform · Documented data/JSON export and workload controls

RANK

Evaluate a bounded workflow after confirming plan, credits, allowed data and output ownership; a compliance badge is not a permission to upload everything.

Pros: Knowledge and actions in one documented workspace · Credit, role and workflow administration

RANK

Consider the exact model and contract, then validate compatibility and cost; the published speed pitch is not an independent performance result.

Pros: Documented SDKs and compatibility differences · Published separation of content and usage-data handling

RANK

Check the exact node/model and operating channel first; price its credits, GPU and storage and retain your own required outputs.

Pros: Documented hosted workflow and API paths · Separate credit and retention documentation

RANK

Map seat, credit and connector permissions together; obtain hosted terms before placing sensitive or commercially restricted content in a workspace.

Pros: Documented agent access and configuration · Separate seat and usage controls

RANK

Choose the operating channel and persistence policy first, then budget sandbox seconds and model services separately.

Pros: Documented machine, code and persistence APIs · Apache code and distinct hosting options

RANK

Choose the search mode and content units before estimating cost, and settle query-training and downstream content permissions separately.

Pros: Documented retrieval filters and response options · Explicit search and content metering

RANK

Configure approval gates and the data tier first, then budget the core components plus optional charges; a subscription does not make actions or training rights unrestricted.

Pros: Documented connector approvals and triggers · Detailed usage and BYOK cost components

RANK

Choose hosted maintenance mode or operated code deliberately, and price observability, stored logs and model calls separately.

Pros: Documented routing and tracing options · Apache repository supports a separate self-hosting path

RANK

Cheap glue for the home and the last mile of consumer habits — a wire between services, not a workflow engine to grow into.

Pros: A recipe for the lamp, not the CRM · Huge device catalog

RANK

A single-file local model runner with roleplay heritage and an AGPL licence — pick it for the character front ends, and watch the upstream project for inference features.

Pros: Unzip and run, genuinely one file · GPU layers without toolchain dances

RANK

Open-source labelling for any modality, including agent traces — the tool handles the interface, and the agreement problem stays yours.

Pros: Configurable labelling interface, so a new task does not mean a new tool · Open source and self-hostable, which matters when the data is sensitive

RANK

Retrieval without a server — LanceDB beside the data, Qdrant in front of it.

Pros: Zero infrastructure to operate · Cheap object-storage scaling

RANK

European LLM observability with agents as the target — buy it for residency and procurement documents, not because the free tier is generous.

Pros: EU data residency by default · Tracing, evals, and guardrails together

RANK

An AI teammate hired for a recurring job, not a canvas to draw on — good for taking chores off a small team, metered in credits as it thinks.

Pros: Templates for common ops roles · Learns from examples, not specs

RANK

OpenAI's API shape on hardware you own — the local gateway for stacks that already speak the SDK.

Pros: OpenAI SDK keeps working against localhost · More than chat: images, audio, embeddings

RANK

The default open-source choice for tracking and the ML lifecycle, with the cost moved from licence fees to your own operations.

Pros: Open source and self-hostable, so the run history and data stay under your control · No per-seat cost, which matters at team scale

RANK

OpenRouter is the pragmatic default for teams that want model choice as an operational property, not a rewrite.

Pros: One API for hundreds of models · Live price/latency leaderboards

RANK

The evaluation platform this entry described is no longer what the company sells — it now builds training tasks and simulated environments for frontier laboratories, on enquiry only.

Pros: Judge LLMs tuned to your domain · Copyright detection built in

RANK

The managed retrieval layer of production AI — rent the boring parts, ship the product.

Pros: Zero-ops vector search at billion scale · Serverless pricing now matches usage

RANK

Python as a cluster — Ray to distribute the functions, Kubernetes to schedule the containers.

Pros: From laptop to cluster without rewrites · Serve, train, and data in one runtime

RANK

The unit-economics breaker of image generation — when cost per image is the product, Runware wins.

Pros: Lowest credible per-image cost in the industry · Sub-second latency at scale

RANK

Consider SerpApi when an API-based search-data workflow fits your needs; verify quota, source rights and privacy settings before production use.

Pros: Documented controls and structured output · Successful-search accounting and cache rules are explicit

RANK

Throughput through reuse — SGLang when prefixes repeat, vLLM when breadth rules.

Pros: Prefix reuse slashes repeated compute · Benchmark-topping throughput

RANK

The pragmatic default for “give my model the web” — boring, predictable, everywhere.

Pros: Dead-simple integration into agent frameworks · Results arrive scraped, chunked and ranked

RANK

The lighter open console — ToolJet if internal should start small, Appsmith if it should have the bigger community.

Pros: A lighter open start · Self-host the panel

RANK

NVIDIA's multi-backend model server — the consolidation layer for mixed vision, speech and language estates, and configuration work before the first request goes through.

Pros: Squeezes GPUs batching can reach · One server, many frameworks

RANK

The Twilio of the voice-agent era — assemble a phone AI in a week instead of a year.

Pros: Skips months of telephony audio engineering · Every component swappable — no lock-in to one model

RANK

Enterprise Gemini next to BigQuery — pick it when the warehouse is already Google Cloud.

Pros: Gemini next to BigQuery under one IAM · Tuning and RAG without a second vendor

RANK

The conversation-design canvas that outlived the chatbot winter — default when CX owns the agent.

Pros: Canvas CX and product can both edit · Knowledge grounding without a custom RAG app

RANK

The default experiment tracker for teams serious about ML, with the caveat that its owner now sells the compute it runs on and its domain is changing.

Pros: Turns scattered training runs into a comparable record · Sweeps and dataset visualisation in the same place as the logs

RANK

Agent observability inside the Weights & Biases platform — buy it if training experiments and production behaviour should share one vendor and one bill.

Pros: Training and inference in one lineage · Evals beside experiment tracking

RANK

The open-source retrieval workhorse — control and hybrid search without the managed bill.

Pros: Open source core you can run anywhere · Hybrid search solves rare-term retrieval

RANK

A lower-priced model family with an OpenAI-compatible door and an explicit coding subscription — the developer documentation is the real product page, so read that rather than the chat.

Pros: Agentic coding out of the box · Pricing that undercuts the majors

RANK

Tracing plus evaluation for agents, with an open-source path into it — most useful once your application has more than one step.

Pros: Tracing and evaluation together, which is what diagnosing an agent actually needs · An open-source observability option, so instrumentation is not tied to the paid tier

RANK

Hands you the whole stack, database and login included — the thing to check before committing is what leaving with the code looks like.

Pros: Auth and database preinstalled · Apps usable by real users on day one

RANK

Consider Baseten when managed inference fits your operating model. Compare full lifecycle charges and the required hosting contract before choosing a deployment.

Pros: Hosted APIs and custom deployments under one platform · Documented scaling and operational controls

RANK

The engineer-friendly agent studio with a self-host door — pick it when Voiceflow is too closed.

Pros: Canvas plus real code when you need it · Self-host path for regulated chat

RANK

The independent option in the AI search stack — own index, no tracking, honest pricing.

Pros: Independent index — real alternative data source · Privacy-first with no user tracking

RANK

The enterprise LLM vendor that treats your data boundary as the product requirement — not an afterthought.

Pros: Deployment freedom: your cloud, or fully on-prem · Embed/Rerank stack is production-proven at scale

RANK

The integration and permission layer for agents — most valuable once your agent needs to act inside accounts it does not own.

Pros: Delegated, scoped auth is the hard part of agent integrations and it is the product here · Sandboxed execution and org-level governance for deploying beyond a prototype

RANK

AI-native cloud infrastructure that now also owns the experiment tracker — convenient if you want one vendor, and a concentration to weigh if you do not.

Pros: Purpose-built for accelerators rather than retrofitted general-purpose cloud · Compute, Kubernetes and storage sold together, which is how training actually gets provisioned

RANK

DeepInfra is the pragmatic utility bill for open-model inference — broad, cheap and boringly reliable.

Pros: Very broad open-model catalog · One API shape for text, vision and audio

RANK

Open-source evaluation and drift monitoring for both classic ML and LLM systems, aimed at the failures that raise no error.

Pros: Open source with a hosted option, so the library can be tried before anything is bought · Covers data drift and LLM evaluation together for systems that contain both

RANK

RAG with the retrieval visible and self-hostable beside your own models — read the non-standard licence before offering it as a service.

Pros: Retrieval debugging as a first-class UI · Self-hosted with your own models

RANK

Fireworks AI is where open models go when throughput and reliability become product requirements.

Pros: Top-tier serving throughput · Fine-tuning and LoRA composition

RANK

The original one-click local LLM app — still the Nomic door, especially for offline file RAG.

Pros: Installer, not a build from source · Offline RAG over local files

RANK

A GPU cloud where the rate card is only the start of the cost question — availability and egress terms decide the rest.

Pros: Offers both on-demand and reserved capacity, so the commitment level can match the workload · Managed inference on the same account as training capacity

RANK

The observability habit every production LLM team needs — Smith is where most start.

Pros: Sees exactly where chains and agents fail · Evals wired to real production traces

RANK

Document parsing sold as a service, with an open-source local parser as the alternative — buy the hosted tier when volume and mess are the problem, not when data must stay in your network.

Pros: Parses tables, forms, and layouts · Prototype-to-production with the same API

RANK

Consider the current Marqo service for commerce discovery after reviewing data and commercial terms. Treat the deprecated open-source engine as a separate maintenance and migration decision.

Pros: Commerce search and merchandising in one documented platform · Catalog and shopper-event integration paths

RANK

The Eastern twin of Hugging Face — mandatory browsing for Qwen-ecosystem work.

Pros: First home of Qwen releases and Chinese-community models · Free notebook inference without local hardware

RANK

An AI cloud whose differentiator is the platform around the GPUs, not the GPUs — test time-to-cluster against your own schedule.

Pros: Covers training and serving rather than only renting capacity · Managed inference and MLOps included, which changes how much you build yourself

RANK

A live-data framework from a company that now leads with frontier research instead — still the right tool when a retrieval index must follow its sources, but check the corporate front page before assuming what you are buying.

Pros: Indexes that never go stale · One framework from source to serving

RANK

The shortest way to a cited, current answer from an API — budget for the searches as well as the tokens.

Pros: Search-grounded answers with citations from one API call · Four tiers with per-million-token prices published

RANK

An open-source RAG engine that begins where retrieval quality is actually decided — the parsing — rather than at the vector index.

Pros: Starts from document parsing, which is where most retrieval quality is actually decided · Open source and self-hostable

RANK

The engineer-owned assistant framework — pick it when Voiceflow is too closed and a widget is too small.

Pros: Flows you can test, plus an LLM · Self-host path for regulated assistants

RANK

Teams of narrow, role-specific agents sold with governance and a deployment partner — an enterprise programme rather than a cheap single automation.

Pros: Teams of agents, not one bot · Humans review the critical steps

RANK

The pragmatic way to ship open models as APIs — start here before you build GPU infrastructure.

Pros: One-line API access to thousands of open models · Per-second billing with scale-to-zero — no idle costs

RANK

Consider Runpod when you can operate container workloads. Compare Pod and Serverless lifecycle costs, storage persistence and the chosen provider before deployment.

Pros: Separate persistent and request-serving workflows · Custom containers and documented storage choices

RANK

Human data and evaluation for organisations running a serious model programme — a vendor contract, not a product you try out.

Pros: Human labelling and evaluation that a model cannot do for itself · Covers the full data pipeline rather than one labelling job

RANK

Google results at agent-scale prices — the quiet workhorse under many research agents.

Pros: Cheapest serious way to get Google SERP data · Millisecond latency fits interactive agents

RANK

LLMs as warehouse functions — the path of least governance friction when the table cannot leave Snowflake.

Pros: LLM where the governed table already sits · SQL-shaped, so analysts can call it

RANK

An agent builder aimed at IT governance in regulated industries — now owned by Asana, which the vendor says changes nothing about the product, and which no third-party statement read here corroborates.

Pros: Audit trails built for enterprises · Permissions across teams and agents

RANK

No-code phone agents — the canvas when CX owns the call and engineering will not.

Pros: CX can edit the flow without an API team · Numbers and minutes in one product

RANK

Hugging Face's model server, now archived and in maintenance mode — use it if you already run it, and choose vLLM or SGLang for anything new.

Pros: Hub models served in minutes · Continuous batching for throughput

RANK

Formerly Text Generation WebUI, now TextGen: a desktop app that puts many local engines, character chat, adapters, image generation and an OpenAI-compatible API behind one window.

Pros: Every backend behind one UI · Character chat and notebook modes

RANK

Open-source fine-tuning whose point is fitting on the GPU you have — plus a desktop app, new enough that its maturity is still an open question.

Pros: Memory efficiency decides whether a fine-tune fits on hardware you already have · Covers text, image, speech and embedding models rather than only text

RANK

A language-to-SQL agent rebuilt around user identity and permissions — but the famous repository is archived, so make sure you are following the new version's documentation, not the old one's.

Pros: Accuracy compounds with your examples · Open source, bring any LLM

RANK

Industrial-grade retrieval — Vespa for the heavy lifting, Marqo for the bundle.

Pros: Ranking logic as expressive code · Inference inside the query path

RANK

A self-hosted model server that treats serving as a service, not a binary — pick it for an internal platform across several model types, not for one laptop.

Pros: Model zoo beyond just LLMs · Cluster mode out of the box

RANK

A governed, dated body of context shared by many agents, sold to enterprises — the graph engine underneath is open source, but that is a different product with different documentation.

Pros: Facts carry validity intervals · Contradictions resolve instead of piling

RANK

A database for AI data that keeps embeddings and metadata together — the improvement claims need testing against your own corpus, not the page.

Pros: Treats embeddings and their metadata as one dataset, which is how retrieval quality is actually decided · Aimed at agent workloads rather than at conventional applications

RANK

Managed Ray for the parallel half of AI work — worth it if cluster operations are the thing standing between you and throughput.

Pros: Managed operations for Ray, which is powerful and operationally fiddly to run yourself · Runs in your own cloud account rather than only on the vendor's

RANK

Fine-tuning driven by a config file rather than a training script — reproducible and quick to start, and not a substitute for good data.

Pros: Config-file training makes runs reproducible and diffable · Ready-made recipes for the setups people actually run

RANK

An infrastructure control plane whose case rests on utilisation — measure yours before and after, since definitions vary by vendor.

Pros: Addresses utilisation, which is where GPU money is actually lost · Covers both the infrastructure control plane and the development tooling above it

RANK

Agents inside the Microsoft tenant — default when Teams is the workplace, skip it if it is not.

Pros: Lives where the users already are: Teams · Graph and Dataverse grounding without a new RAG app

RANK

Security for AI systems framed around behaviour rather than permissions, with the usual caveat that the guardrail is a model too.

Pros: Treats behaviour rather than access as the security boundary, which is where agents actually go wrong · Covers both red teaming and runtime defence, which answer different questions

RANK

Model routing that pays off on the distribution of real traffic — and needs a quality floor so a misjudged request does not cost more than it saves.

Pros: Targets the fact that most traffic does not need the most expensive model · Per-request decisions rather than a fixed provider choice

RANK

Open weights, production packaging — OpenLLM to ship, Ollama to run at your desk.

Pros: From weights to endpoint in one command · Versioned, reproducible Bentos

RANK

A peer-to-peer way to run models too large for your own machine — good for experiments and study, wrong for anything that has to answer promptly, and the repository has been quiet since 2024.

Pros: 70B-class models without owning GPUs · Fine-tuning on the public swarm

RANK

Prompts sold one at a time, with previews as the proof — but the site refused every fetch this time, so treat this entry as an unverified description rather than a current one.

Pros: Output previews prove the prompt works · Dollars, not subscriptions

RANK

Moves the prompt out of the repository so it can be iterated and tested by the people who own the wording, with the responsibility staying with them.

Pros: Lets people who do not write the application iterate on prompts · Evaluations run before a change ships rather than after users see it

RANK

Evaluation aimed squarely at how RAG actually fails, with the caveat that some of the judging is done by a model.

Pros: Metrics designed for retrieval failures rather than generic text similarity · Synthetic test data means you can evaluate before you have real traffic

RANK

Domestic hosted Comfy — the cloud when the workflow language is already Chinese.

Pros: Comfy graphs without a local GPU · Templates in the language the community uses

RANK

A vendor whose centre of gravity is on-premises AI systems rather than self-serve API calls — check which half you actually need.

Pros: Sells both an API service and on-premises systems, so the same vendor can cover either shape of deployment · Own silicon rather than resold capacity

RANK

Together AI is the de-facto home for serious open-weight deployment — research credibility plus production plumbing.

Pros: Huge catalog of open models served fast · Fine-tuning into custom endpoints

RANK

A vector search engine that trades latency for storage-priced scale — check the latency against your own access pattern before the cost saving.

Pros: Prices a large, unevenly accessed corpus at storage rates instead of keeping all of it resident · Full-text and vector search together rather than only embeddings

RANK

Enterprise retrieval and agents whose selling point is where they can run, not how they score.

Pros: Deployment options that include air-gapped on premises, which decides the shortlist for regulated buyers · Retrieval, agents and governance in one offering rather than assembled from parts

RANK

A model-routing and tuning shop now rather than a frontier model vendor — evaluated on your own traffic and your own evals, or not at all.

Pros: Attacks token cost on your own traffic rather than with a published rate comparison · Gateway endpoint is drop-in, so it does not demand a rewrite

RANK

Not the smartest AI — the most trustworthy delivery of it, at a scale nobody else has.

Pros: Privacy architecture you can actually verify · AI woven into the OS, not another app

RANK

A validation layer for LLM I/O that earns its place by deleting defensive parsing code — and that cannot decide your policy for you.

Pros: Removes hand-written defensive parsing rather than adding to it · Open source, so validators can be inspected and extended

RANK

Novita AI is a budget-friendly utility cloud for open models across text, image and audio.

Pros: Hundreds of open models · Per-token and per-image pricing

RANK

A sensible way to stop hard-coding one model vendor — with the caveat that the product is mid-acquisition and a gateway sees everything you send it.

Pros: Provider swaps become configuration instead of a code change · Logging, guardrails and prompt management in one control plane

RANK

SiliconFlow is the low-cost gateway to China's leading open models.

Pros: Leading Chinese open models · Standardized APIs

RANK

Consider Vast.ai when you can choose and operate a provider-backed instance. Compare total compute, storage and transfer costs with interruption and security requirements.

Pros: Selectable machine and rental configurations · Documented host and access-type distinctions