Kognic

Kognic, founded in 2018 in Gothenburg, Sweden and formerly known as Annotell, operates an annotation platform and related workforce services for developers of autonomous driving and ADAS systems. The platform handles camera, LiDAR and radar data in a single multi-sensor workspace, supporting perception, planning and end-to-end model training. Annotation types include 3D cuboids, multi-LiDAR cuboids, polylines, polylanes, semantic segmentation, points and curves, plus 2D bounding boxes, polygons, points, curves and text. Use cases span lane and road-marking labeling, dynamic and static object annotation, instance and 3D segmentation, freespace, traffic sign and traffic light recognition, parking spot detection, camera blockage, light source detection, interior sensing and driver monitoring.

The product combines human annotators with automation: pre-annotation import from customer or third-party models, one-click cuboid creation, smart interpolation with ego-motion compensation, intelligent propagation across frames, Segment Anything model integration, multi-sensor projections and automatic static quality checks. More than 90 automated "Checker App" validators flag annotation issues, and productivity metrics track throughput, review acceptance and time per annotation. The platform is API-first for pushing sensor data, submitting pre-annotations and exporting production output; inputs include PNG, JPG, JPEG, WEBP and AVIF images and PCD, LAS and CSV point clouds, with OpenLABEL and custom export converters. A Language Grounding capability, built on a "Chain of Causation" methodology with Write, Edit and Rank modes, produces structured reasoning data for vision-language and visual-language-action models.

The company describes itself as extending human-in-the-loop annotation beyond routine labeling into trajectory ranking, intent judgment, multimodal grounding and alignment tasks. Enterprise features include multi-account workspace management, role-based access and task allocation, sandbox mode, custom workflows, a taxonomy editor, and ISO and TISAX Level 3 certifications.

Founding story

Kognic was founded in 2018 in Gothenburg, Sweden by engineering physicists Daniel Langkilde and Oscar Petersson, who drew on their work in deep learning and set out to build annotation software that they felt did not yet exist. The stated founding mission was to help autonomous systems learn faster through scalable, cost-efficient human feedback.

Business model

Kognic sells access to its annotation platform and associated annotation labor under three deployment models: a managed service in which Kognic's platform and expert annotation partners deliver the work, a platform license in which customers deploy the tools within their own workflows and internal teams, and a hybrid model combining the platform with the customer's preferred BPO or workforce provider. Enterprise onboarding, specialized training and expert guidance are offered alongside the software.

Revenue derives from platform licensing and delivered annotation services; the company states it measures success by customer appreciation and revenue rather than fundraising and, as of the 2024 round, said it was on track to return to profitability by 2025.

Traction

More than 100 million annotations delivered across over 120 projects, with customers including Qualcomm, BMW, Zenseact, Continental and Bosch. The company reported 90% year-over-year growth around its 2024 round. Published customer outcomes include Zenseact saving 10.5 MSEK and up to 47% cost savings on dynamic object annotation through pre-label workflows, and claims of up to 68% annotation time saved using auto-labels and up to 3x faster processing.

Latest developments

In 2026 Kognic published platform updates including redesigned task performance metrics, Studies for tracking group improvement, quality-based review sampling, geometry-only review edits and faster search, announced a partnership with Mobilix to serve the Korean market for robo-buses, robo-taxis and Level 2+ ADAS, and marked passing 100 million delivered annotations. Product focus has expanded toward Language Grounding and reasoning-oriented annotation for vision-language and visual-language-action models.