Platforms & Infrastructure

About this category

83 products from 80 companies, the earliest first recorded in 2016.

By business model: Freemium 40 · Open (open weights/source) 24 · Paid 18 · Free 1.

Recorded uses in this category: Running models yourself, Automation agents, Data analysis, Questions over your own documents, Search, Editing an existing codebase.

The longest-standing record here is Hugging Face (2016); the most recent arrival is Gcore (2025).

This category covers hubs and infrastructure for hosting, sharing, and running AI models and datasets. They provide the APIs, compute, and tooling that AI applications build on.

Fields taken from the product records use the record dates shown. Sourced claims are labelled per product: verified against the linked page on the date shown, taken from the product record without re-checking, or not independently verified.

Every product here, one line each

• Agenta — Freemium; Automation agents. Made by Agenta, first recorded Not recorded. Pricing: Hobby gives 75 daily free credits on up to 10 days per month, plus a one-time 500-credit signup bonus. Starter is USD 29 per month with 2900 credits; credits cover included models and runtime, and paid plans do not charge per seat.
• Agno — Open (open weights/source); Automation agents. Made by Agno, first recorded Not recorded. Pricing: Self-hosting the open-source project and buying the hosted platform are separate choices. The hosted service has its own plans and usage limits; self-hosting still requires infrastructure and any model-provider charges.
• Aleph Alpha — Paid; Running models yourself. Made by Aleph Alpha, first recorded 2019. Pricing: Aleph Alpha sells to organizations rather than consumers and does not publish list prices; commercial terms are arranged directly with enterprise and public-sector customers.
• Amazon Bedrock — Paid; Running models yourself. Made by Amazon Web Services, first recorded 2023. Pricing: Pricing is usage-based per input/output token, with on-demand and provisioned-throughput options.
• Amazon Nova — Paid; Running models yourself. Made by Amazon Web Services, first recorded 2024. Pricing: Amazon Nova is billed through AWS, typically via Amazon Bedrock, with usage-based pricing that varies by model and service.
• Arize Phoenix — Free; Data analysis. Made by Arize AI, first recorded Not recorded. Pricing: Phoenix can be run on your infrastructure under Elastic License 2.0; its source availability is not an unrestricted open-source license. Infrastructure and evaluator-model costs are separate, and managed Arize offerings have separate commercial terms.
• Baseten — Paid; Running models yourself. Made by Baseten, first recorded 2019. Pricing: Dedicated GPU compute is billed per minute with no charge for idle time, model APIs are billed per million tokens, and new accounts come with credits.
• BentoML — Open (open weights/source); Running models yourself. Made by BentoML, first recorded Not recorded. Pricing: Self-hosting the open-source project and buying the hosted platform are separate choices. The hosted service has its own plans and usage limits; self-hosting still requires infrastructure and any model-provider charges.
• Braintrust — Freemium; Data analysis. Made by Braintrust, first recorded Not recorded. Pricing: Free has a USD 0 platform fee, with 1 GB of processed data and 10000 scores per month before metered overages. Pro is USD 249 per month; model credits, data processing, scores and retention are separate billing dimensions.
• Browser Use — Freemium; Automation agents. Made by Browser Use, first recorded Not recorded. Pricing: The Python library and cloud service are separate. Cloud browsers cost USD 0.02 per hour plus proxy traffic; hosted agents add a 20% model surcharge plus browser and traffic costs. Eligible signups receive USD 1 of introductory credit.
• Browserbase — Freemium; Automation agents. Made by Browserbase, first recorded Not recorded. Pricing: A free plan is available. Developer is USD 20 per month with 100 browser hours; browser overages, proxies, search, fetch and model usage have separate meters, so the monthly fee does not cover unlimited automation.
• Cerebras — Freemium; Running models yourself. Made by Cerebras Systems, first recorded 2024. Pricing: Inference is usage-priced per token and starts with a free trial; training systems and on-premises deployments are sold separately.
• Chroma — Freemium; Search. Made by Chroma, first recorded Not recorded. Pricing: The open-source database can run locally. Cloud Starter has a USD 0 monthly platform fee with USD 5 credits; writes, stored data, queried data and returned network traffic are metered separately, so zero base fee does not mean unlimited cloud usage.
• Civitai — Freemium; Running models yourself. Made by Civitai, first recorded 2022. Pricing: Browsing and downloads are free; on-site generation and training consume Buzz credits, available through memberships or one-off purchases.
• CoreWeave Forge — Freemium; Data analysis, Running models yourself. Made by CoreWeave, first recorded Not recorded. Pricing: Free supports personal development with limited storage and ingestion. Pro starts at USD 60 per month billed monthly; storage, Weave ingestion, inference and sandbox compute have separate meters. Academic access has eligibility requirements, not a blanket commercial exemption.
• Daytona — Freemium; Automation agents. Made by Daytona, first recorded Not recorded. Pricing: Daytona bills sandbox compute, memory and storage by usage, with USD 200 of introductory compute credit. CPU and GPU resources have different rates; introductory credit is not a recurring free allocation.
• DeepInfra — Paid; Running models yourself. Made by DeepInfra, first recorded Not recorded. Pricing: Pricing depends on the model: supported language models use token billing while other workloads may use execution-time billing. Private model deployments and GPU rental are distinct from shared inference API rates; there is no single monthly price for all models.
• Dify — Open (open weights/source); Running models yourself, Questions over your own documents. Made by LangGenius, first recorded 2023. Pricing: Self-hosting Dify is free under its open-source license; Dify Cloud offers a free sandbox and paid plans, with enterprise licensing for advanced needs.
• Docling — Open (open weights/source); Questions over your own documents. Made by Docling, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• DSPy — Open (open weights/source); Automation agents. Made by Stanford NLP, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• E2B — Freemium; Automation agents. Made by E2B, first recorded Not recorded. Pricing: Hobby has USD 100 in one-time credits, followed by metered usage. Pro adds a USD 150 monthly base fee plus CPU, RAM and storage usage; the base fee is not an unlimited compute subscription.
• Evidently AI — Open (open weights/source); Data analysis. Made by Evidently AI, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• Fal.ai — Paid; Running models yourself. Made by fal, first recorded 2023. Pricing: Pricing is usage-based per inference or compute second; new developers receive free credits.
• Firecrawl — Freemium; Search, Questions over your own documents. Made by Firecrawl, first recorded Not recorded. Pricing: The hosted service offers 1000 free credits per month. Paid subscriptions and extra credits cover metered search or scraping; the advertised USD 16 monthly entry price is billed yearly, not a monthly-billed USD 16 plan.
• Fireworks AI — Paid; Running models yourself. Made by Fireworks AI, first recorded 2023. Pricing: Pricing is usage-based per token, with dedicated capacity options; new developers receive free credits.
• Gcore — Freemium; Running models yourself. Made by G-Core Labs S.A., first recorded 2025. Pricing: Gcore uses usage-based pricing with volume discounts. Some products, such as CDN, FastEdge, and DNS, offer free tiers, while AI and GPU usage is paid; H100 GPU virtual machines are billed per second and billing pauses when they are powered off.
• Gemini Enterprise Agent Platform — Paid; Running models yourself. Made by Google, first recorded 2021. Pricing: Pricing is usage-based through Google Cloud, varying by model, training, and serving resources; new customers receive free credits.
• Giskard — Freemium; Data analysis. Made by Giskard, first recorded Not recorded. Pricing: Giskard offers open-source testing components and the commercial Giskard Hub. Hub and enterprise security evaluations are arranged with the vendor; a limited audit promotion is not an ongoing free hosted plan, and test-generation model costs may be separate.
• Groq — Freemium; Running models yourself. Made by Groq, first recorded 2024. Pricing: A free tier is available; production usage is priced per token through Groq's API.
• Guardrails AI — Open (open weights/source); Data analysis. Made by Guardrails AI, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• Guidance — Open (open weights/source); Data analysis. Made by Guidance contributors, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• Haystack — Open (open weights/source); Automation agents, Questions over your own documents. Made by deepset, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• Helicone — Freemium; Data analysis. Made by Helicone, first recorded Not recorded. Pricing: Hobby includes 10000 requests per month, 1 GB storage and one seat. Pro costs USD 79 per month plus usage; request, storage and retention limits should be compared separately from upstream model API charges.
• Hugging Face — Freemium; Running models yourself. Made by Hugging Face, Inc., first recorded 2016. Pricing: Public hosting and downloads are free; paid plans cover compute, inference endpoints, and enterprise features.
• Infomaniak AI Services — Freemium; Running models yourself. Made by Infomaniak Network SA, first recorded 2024. Pricing: Credit-based and usage-billed: the vendor's support guide says each request consumes a credit or LLM tokens, that one million free credits cover a first month without commitment, and that pricing is shown on the sales page; a credit card is required and a spending limit can be set. No flat plan price or per-token table was readable on the pages fetched, so this record sets no comparable paid_from figure.
• Jan — Open (open weights/source); Running models yourself, Questions over your own documents. Made by Menlo Research, first recorded 2023. Pricing: Jan is free and open source; users optionally pay third-party providers if they connect cloud models.
• Jina AI — Freemium; Search, Questions over your own documents. Made by Jina AI, first recorded Not recorded. Pricing: Reader allows keyless access with lower rate limits; an API key increases limits. Embedding, reranking and Reader usage have their own token and service terms, so a Reader allowance should not be treated as free access to every Jina model.
• LanceDB — Open (open weights/source); Search, Data analysis. Made by LanceDB, first recorded Not recorded. Pricing: Self-hosting the open-source project and buying the hosted platform are separate choices. The hosted service has its own plans and usage limits; self-hosting still requires infrastructure and any model-provider charges.
• Langfuse — Freemium; Data analysis. Made by Langfuse, first recorded Not recorded. Pricing: The open-source platform can be self-hosted. Cloud has a Hobby tier; Core is USD 29 per month with 100000 units included and additional units metered. Cloud retention and team features vary by tier, and external model evaluation costs remain separate.
• Letta — Freemium; Automation agents. Made by Letta, first recorded Not recorded. Pricing: Free supports up to 3 stateful agents with your own model access. Pro is USD 20 per month; developer API plans have separate base and usage charges, including active agents and tool execution. A personal plan is not an unlimited API entitlement.
• LiteLLM — Open (open weights/source); Automation agents. Made by LiteLLM, first recorded Not recorded. Pricing: The open-source gateway is free to self-host. Enterprise is licensed by annual request capacity, architecture and support rather than per model token; underlying model-provider charges and self-hosted infrastructure remain separate.
• LM Studio — Freemium; Running models yourself. Made by LM Studio, first recorded 2023. Pricing: Checked 2026-11-02: Free covers local models and Bionic Agent. Bionic+ is USD 20/month and Pro USD 100/month for optional cloud usage allowances; cloud credits cover use beyond the included allowance. These paid individual tiers are available now; the page separately says team subscription plans are coming soon.
• Marker — Open (open weights/source); Questions over your own documents. Made by Datalab, first recorded Not recorded. Pricing: Marker code is Apache-2.0, but its model weights use modified AI Pubs Open Rail-M terms: research, personal use and businesses below USD 5 million funding/revenue qualify for free use; larger commercial use needs a license. Datalab’s hosted API is a separate metered service.
• Mastra — Open (open weights/source); Automation agents. Made by Mastra, first recorded Not recorded. Pricing: Self-hosting the open-source project and buying the hosted platform are separate choices. The hosted service has its own plans and usage limits; self-hosting still requires infrastructure and any model-provider charges.
• Mem0 — Freemium; Automation agents. Made by Mem0, first recorded Not recorded. Pricing: Hosted Free covers 10000 adds and 1000 retrievals with model calls during the first 30 days. Using your own LLM key raises the listed allowance to 50000 adds and 10000 retrievals per month; overages are metered with a user-set monthly cap, and larger contracts are custom.
• Microsoft Foundry — Paid; Running models yourself. Made by Microsoft, first recorded 2024. Pricing: Pricing is usage-based through Azure, varying by model and service; many tools have no separate platform fee.
• Milvus — Open (open weights/source); Search. Made by Zilliz, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• MindsHub — Freemium; Automation agents, Editing an existing codebase. Made by MindsDB, first recorded Not recorded. Pricing: Hosted MindsHub advertises ongoing free access to its Air model under fair-use limits. A topped-up balance pays for other models; team budgets and spend controls do not make premium-model calls free. Running the source stack yourself has separate infrastructure and provider costs.
• Modal — Freemium; Running models yourself. Made by Modal Labs, first recorded 2022. Pricing: Pricing is usage-based by compute consumed, with a monthly allowance of free compute included.
• Nebius — Paid; Running models yourself. Made by Nebius Group N.V., first recorded 2024. Pricing: Pricing is usage-based: GPU compute per GPU-hour, plus storage and networking, with token-based inference on Token Factory. Some services are listed as free and promo credits are available, but there is no general free tier.
• Novita AI — Paid; Running models yourself, Automation agents. Made by Novita AI, first recorded Not recorded. Pricing: Model APIs are metered by model-specific input, output and cached-token rates. GPU Cloud and Agent Sandbox have separate resource pricing; verify account credit and the chosen product’s rate rather than treating all services as one subscription.
• Nscale — Paid; Running models yourself. Made by Nscale Limited, first recorded 2025. Pricing: Nscale does not publish a pricing page. Serverless inference is billed pay-as-you-go per token or request, compute is also available through reservations, and new accounts receive a small free credit according to the documentation.
• NVIDIA Dynamo-Triton — Open (open weights/source); Running models yourself. Made by NVIDIA, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• NVIDIA NeMo — Freemium; Running models yourself, Data analysis. Made by NVIDIA, first recorded Not recorded. Pricing: NeMo includes open-source frameworks and separately offered enterprise microservices. Training compute, storage, model weights and enterprise support have their own costs and licenses; an open framework license does not make every NeMo service or model free.
• NVIDIA NIM — Freemium; Running models yourself. Made by NVIDIA, first recorded Not recorded. Pricing: Developer Program access covers NIM endpoints and containers for development and testing. Production security updates and support are associated with NVIDIA AI Enterprise or partner endpoints; GPU infrastructure and model-specific terms remain separate from prototype access.
• Ollama — Open (open weights/source); Running models yourself. Made by Ollama, first recorded 2023. Pricing: Ollama is free and open source; optional hosted services are offered separately.
• Open WebUI — Open (open weights/source); Running models yourself. Made by Open WebUI, first recorded 2023. Pricing: Open WebUI is free and open source to self-host; an enterprise plan adds custom theming, SLAs, and support for a fee.
• OpenRouter — Paid; Running models yourself. Made by OpenRouter, first recorded 2023. Pricing: OpenRouter uses pay-as-you-go pricing that passes through provider rates, with enterprise plans; several models can be used free, and zero-token responses are not charged.
• Opik — Freemium; Data analysis. Made by Comet, first recorded Not recorded. Pricing: The self-hosted open-source edition is free. Free Cloud includes 25000 spans per month and 60-day retention; Pro Cloud is USD 19 per month with 100000 spans. Evaluator-model charges and enterprise terms remain separate.
• Outlines — Open (open weights/source); Data analysis. Made by .txt, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• OVHcloud AI Endpoints — Freemium; Running models yourself. Made by OVH Groupe SA, first recorded 2025. Pricing: Billed by usage rather than by subscription: each model's price per million tokens is published in the public catalogue, ranging from around €0.01 for an embedding model to about €0.91 for a large visual model, with a few models (moderation and image generation) listed as free. OVHcloud offers a US$200 Public Cloud free-trial credit to first-time projects. Rates change, so the catalogue is the source of record; this record sets no comparable paid_from figure.
• Pinecone — Freemium; Running models yourself. Made by Pinecone, first recorded 2021. Pricing: The Starter tier is free with capacity limits; Standard and Enterprise plans bill by storage, reads, and writes on serverless usage.
• Poe — Freemium; Running models yourself. Made by Quora, first recorded 2023. Pricing: A free tier is available; the paid subscription grants monthly compute points usable across premium models.
• Prisma AIRS AI Gateway — Freemium; Automation agents. Made by Palo Alto Networks, first recorded Not recorded. Pricing: The official Portkey pricing page lists a free developer tier with 10000 recorded logs per month and Production at USD 49 monthly with log overages. The newly branded Prisma AIRS enterprise gateway should be quoted separately; model-provider charges are not included.
• Promptfoo — Freemium; Data analysis. Made by Promptfoo, first recorded Not recorded. Pricing: Community is free and includes up to 10000 red-team probes per month where hosted inference is needed. Enterprise is quoted for team needs; testing your own targets can still incur their model API charges.
• PromptLayer — Freemium; Data analysis. Made by PromptLayer, first recorded Not recorded. Pricing: Free includes 2500 requests and 250 evaluation cells per month. Pro costs USD 49 per month, with transaction overages billed separately; model calls and larger retention or deployment requirements should be budgeted independently.
• PydanticAI — Open (open weights/source); Automation agents. Made by Pydantic, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• Qdrant — Freemium; Search. Made by Qdrant, first recorded Not recorded. Pricing: Open-source Qdrant can be self-hosted. Cloud Free includes 1 GB RAM and 4 GB disk; Standard is resource-based for a dedicated cluster. Embedding-provider costs and larger-cluster charges are separate from the free database tier.
• Replicate — Paid; Running models yourself. Made by Replicate, first recorded 2019. Pricing: Pricing is per second of active compute with no subscription; idle deployments scale to zero and are not billed.
• SambaCloud — Freemium; Running models yourself. Made by SambaNova, first recorded Not recorded. Pricing: SambaCloud separates developer access, production pay-as-you-go and larger subscription-based plans. Rates and rate limits depend on the selected model and plan; API usage is not included merely by installing an OpenAI-compatible client.
• Scaleway — Paid; Running models yourself. Made by Scaleway SAS, first recorded 2024. Pricing: Services are billed by usage; Generative APIs are billed per million tokens on a pay-as-you-go basis or through dedicated deployment, and prices are published on the Scaleway website.
• smolagents — Open (open weights/source); Automation agents. Made by Hugging Face, first recorded Not recorded. Pricing: The open-source software can be run on your own infrastructure. Compute, storage and any external model API usage are separate costs; a managed service is not included.
• STACKIT AI Model Serving — Paid; Running models yourself. Made by Schwarz Digits Cloud GmbH & Co. KG, first recorded 2025. Pricing: Token-based: the product page publishes rates per million tokens by model tier, for example €0.15 for standard input, €0.45 for plus input, €1.50 for premium input, €0.25 for standard output and €1.75 for premium output, plus embedding rates from €0.02, all in the Germany South (EU01) region. No free allowance is advertised on the page read, and this record sets no comparable paid_from figure.
• Steel — Freemium; Automation agents. Made by Steel, first recorded Not recorded. Pricing: Steel Browser can be self-hosted. Steel Cloud Launch has no monthly base fee and gives USD 30 in one-time credits, then usage charges; Scale is USD 250 per month plus usage, with separate browser, proxy and CAPTCHA rates.
• Supermemory — Freemium; Automation agents, Questions over your own documents. Made by Supermemory, first recorded Not recorded. Pricing: Free includes USD 5 of credits renewed monthly. Pro is USD 19 per month with USD 20 included usage; memory and SuperRAG have different meters, and additional consumption is charged separately from the subscription.
• Together AI — Paid; Running models yourself. Made by Together AI, first recorded 2023. Pricing: Pricing is usage-based: per-token for serverless endpoints and per GPU hour for dedicated capacity; free credits are offered for new users.
• turbopuffer — Paid; Search. Made by turbopuffer, first recorded Not recorded. Pricing: Launch has a USD 16 monthly minimum usage charge. Storage, writes, queries and namespace counts contribute to the bill; higher plans add governance and support, so a minimum is not an all-inclusive flat subscription.
• Unstructured — Freemium; Questions over your own documents. Made by Unstructured, first recorded Not recorded. Pricing: Hosted pay-as-you-go starts with 10000 introductory pages and then charges USD 0.015 per page. The starting page allowance is not described as monthly renewal; enterprise deployment and high-volume terms require a quote.
• Upstage — Freemium; Running models yourself, Questions over your own documents. Made by Upstage, first recorded 2020. Pricing: Upstage meters API use by tokens or document pages and offers prepaid tiers (Explore, Build, Scale) that add credits and higher request limits; enterprise pricing is custom. Some features and models have free or beta access, including an education and public-interest program.
• Vespa — Open (open weights/source); Search. Made by Vespa, first recorded Not recorded. Pricing: Vespa can be self-hosted as open-source software. Vespa Cloud is resource-priced using the official calculator and offers a trial; node resources, storage and applicable traffic costs should be assessed separately from free trial access.
• vLLM — Open (open weights/source); Running models yourself. Made by vLLM Project, first recorded 2023. Pricing: vLLM is free and open source under Apache-2.0; costs are limited to the GPU infrastructure it runs on.
• Weaviate — Open (open weights/source); Running models yourself. Made by Weaviate, first recorded 2019. Pricing: Weaviate is free to self-host (BSD-3-Clause); Weaviate Cloud is a paid managed service priced by usage and tier.
• Zep — Freemium; Automation agents. Made by Zep, first recorded Not recorded. Pricing: Free access has credit limits. Flex is USD 125 per month billed monthly with 50000 credits; ingestion and retrieval consume credits and automatic top-ups can add charges. Open-source Graphiti is distinct from the hosted Zep service.

Frequently asked questions