Skip to content

Annotide compared

Six tools that all do data annotation, built on different bets about where your media lives and who hosts the software. This page is a neutral, fact-checked overview; each tool has its own page with sources.

Annotide is self-hosted only: you run it in your own cloud or on your own servers, against your own PostgreSQL and your own object storage. The platform reads and writes media through short-lived signed URLs and never copies it into vendor infrastructure — only annotations are stored, in your database. That design choice shapes every row below.

Annotide CVAT Label Studio Labelbox Encord V7 Darwin
Deployment Self-hosted only (Docker Compose, Helm, Terraform for Azure) Self-hosted (free) or CVAT-hosted cloud; Enterprise self-hosted Self-hosted (free) or HumanSignal-hosted Enterprise/Starter Cloud Hosted SaaS; on-premises/airgapped offered to Enterprise (scope not published) Hosted SaaS; VPC/on-prem at Enterprise tier Hosted SaaS
Where media lives Your own storage (Azure Blob, S3, GCS, local) — never copied to a vendor Your storage if self-hosted; CVAT’s servers for cloud-hosted projects unless you attach your own bucket Your storage if self-hosted; HumanSignal’s infrastructure for cloud plans Labelbox’s cloud (Google Cloud) by default Encord’s cloud by default; “Private Cloud Integration” can keep data in your storage on paid tiers V7’s AWS infrastructure (EU/Ireland per their published data-residency statement)
Licence / source Source-available, Elastic License 2.0; a paid key adds users and Business features Open source, MIT (Community); Enterprise adds proprietary features Open source, Apache 2.0 (Community); Enterprise is proprietary Proprietary, closed source Proprietary, closed source Proprietary, closed source
Starting price Free (Community, up to 3 users, commercial use included); Team from €12/user/month; Business €20/user/month Free (Community, self-hosted); hosted Free/Solo from ~$23–33/mo; Enterprise from ~$12,000/yr Free (Community, self-hosted); Starter Cloud from ~$50/mo Free tier (capped); Starter pay-as-you-go at $0.10/unit No published list price (quote-based) No published list price (quote-based)
SSO OIDC and SAML 2.0, Business and Enterprise (€20/user/month, self-serve) OIDC and SAML, Enterprise only SSO, Enterprise only SSO, add-on for Enterprise accounts SSO, Enterprise tier only SAML / Google OAuth, Enterprise tier
Model-assisted labelling Bring-your-own model gateway: batch pre-labelling, interactive segmentation, active-learning queue ordering, retraining webhook; optional bundled SAM 2.1 image for click-to-polygon segmentation, on CPU (no auto-track or batch auto-annotate) Bundled SAM2/SAM3 auto-annotation (hosted tiers), Hugging Face/Roboflow integrations ML backend interface (bring your own model server) Model-assisted labelling, bring-your-own model, Model product Model-assisted labelling, model evaluation, data agents Auto-Annotate with SAM2, auto-track, “model in the loop”
Media types Image, video, audio, text (NER/relations), PDF, time series, LLM evaluation (ranking/scoring/conversations), multimodal tasks Image, video, 3D point clouds, audio Image, video, audio, text, time series, multimodal Image, video, text, documents, 3D/point cloud Image, video, DICOM/medical, 3D/LiDAR, documents Image, video, PDF/documents, medical (DICOM, SVS)

Sources are listed on each tool’s own page, checked October 2026. Where a vendor doesn’t publish a number (several Enterprise prices, Labelbox’s on-premises scope), the table says so rather than guessing.

You’re a solo researcher, a hobbyist or a team of up to three. CVAT Community or Label Studio Community are free, mature, self-hosted and well documented. Annotide’s Community edition is also free for up to three users, commercial use included, with every annotation feature — useful if you specifically want LLM evaluation or multimodal tasks alongside image/video work in the same tool.

You want a hosted tool and don’t want to run any infrastructure. CVAT Online, Label Studio’s Starter Cloud, Labelbox, Encord and V7 Darwin all offer that; Annotide does not — there is no Annotide-hosted SaaS, by design, so your media never has to leave your own cloud account. If that trade-off (you run Docker/Kubernetes, in exchange for data never touching a vendor) isn’t one you want to make, a hosted tool is the better fit.

Your media can’t leave your own cloud account or on-prem network, for contractual or regulatory reasons. This is Annotide’s core bet: the backend never proxies media, and nothing is copied to R Squared Data Solutions’ infrastructure. CVAT and Label Studio can do this too if you self-host their open-source editions; Encord and Labelbox offer VPC/on-prem options at their Enterprise tiers, at Enterprise pricing.

You need a large, battle-tested open-source community, or a plugin ecosystem. Label Studio has one of the broadest media-type/plugin ecosystems in open-source annotation and a large contributor base. CVAT has a similarly long track record, particularly for computer vision teams and video.

You need a managed annotation workforce, not just software. Labelbox, Encord and V7 Darwin all sell into “give us the data, get back labels” workflows with services attached. Annotide doesn’t offer annotation services — it’s software you or your own annotators run.

You want simple per-seat prices, self-hosted on your own cloud, from an EU vendor. This is where Annotide fits: no usage-based units, every annotation feature free for up to three users, single sign-on, SCIM and quality control at a published €20/user/month instead of a sales call (support is a separate plan), bring-your-own model endpoints with managed identity / service-principal auth, an MCP server for AI coding agents, and LLM evaluation + computer vision annotation in the same product. Weigh that against the real trade-offs: no hosted option, a small vendor (a Finnish sole trader, not a funded company with a large support org), and some gaps noted on the individual pages (the bundled SAM2 does interactive segmentation only, not auto-tracking or batch auto-annotation; the reference training pipeline does boxes but not masks).