https://www.goodailabs.com/ Good AI Labs pursues intelligence that learns from experience and acts in the world through decision models, robot learning, spatial memory and persistent agents. en-GB https://www.goodailabs.com/updates/ https://www.goodailabs.com/updates/Mon, 05 Oct 2026 17:23:32 +0000 New experiments, reports and openings at the lab. https://www.goodailabs.com/blog/reflex-1-fast-decisions/ https://www.goodailabs.com/blog/reflex-1-fast-decisions/Mon, 05 Oct 2026 17:23:32 +0000 A 421M-parameter model for tool routing, intent classification and action selection, with public weights and measured CPU performance. https://www.goodailabs.com/research/reflex-1/ https://www.goodailabs.com/research/reflex-1/Mon, 05 Oct 2026 17:23:32 +0000 A 421M-parameter decision model for agents. One forward pass, on CPU or GPU. https://www.goodailabs.com/blog/samsara-capability-and-compute/ https://www.goodailabs.com/blog/samsara-capability-and-compute/Mon, 05 Oct 2026 05:32:07 +0800 Robot control with memory, in two compact models. https://www.goodailabs.com/research/samsara-vla/ https://www.goodailabs.com/research/samsara-vla/Mon, 05 Oct 2026 05:32:07 +0800 Two compact robot policies for language-conditioned manipulation. Watch them pick, place, open and complete multi-step tasks. https://www.goodailabs.com/blog/a-number-you-cannot-tune-for/ https://www.goodailabs.com/blog/a-number-you-cannot-tune-for/Sat, 03 Oct 2026 12:44:31 +0000 The selected 80,000-step checkpoint finished two episodes ahead of the 34,000-step candidate. All five results from the same 2,000-episode evaluation. https://www.goodailabs.com/blog/re1-video-memory/ https://www.goodailabs.com/blog/re1-video-memory/Sat, 03 Oct 2026 12:44:31 +0000 Search recorded events, check the source video and revisit captured places in 3D. https://www.goodailabs.com/products/re1/ https://www.goodailabs.com/products/re1/Sat, 03 Oct 2026 12:44:31 +0000 Spatial memory for physical AI. Return to an observation, inspect its source video and reconstruct the place for simulation. https://www.goodailabs.com/research/vio-lens-2/ https://www.goodailabs.com/research/vio-lens-2/Sat, 03 Oct 2026 11:58:00 +0000 Research in progress. https://www.goodailabs.com/privacy/ https://www.goodailabs.com/privacy/Sat, 03 Oct 2026 19:00:00 +0800 Website logs and email correspondence. https://www.goodailabs.com/blog/colony-runtime-and-model-budget/ https://www.goodailabs.com/blog/colony-runtime-and-model-budget/Sat, 03 Oct 2026 10:26:42 +0000 Give an agent an ongoing job, a place to work and memory it can use next time. Colony is opening for technical previews. https://www.goodailabs.com/products/colony/ https://www.goodailabs.com/products/colony/Sat, 03 Oct 2026 10:26:42 +0000 Personal agents that keep working across conversations. Give them projects, tools and memory. They can write code, use your Mac, build small apps and return to scheduled work on a computer you control. https://www.goodailabs.com/blog/goodness-first/ https://www.goodailabs.com/blog/goodness-first/Fri, 02 Oct 2026 00:00:00 +0800 We want people to be able to run and adapt the models we build. That puts hardware requirements and operating control in the research brief. https://www.goodailabs.com/terms/ https://www.goodailabs.com/terms/Fri, 02 Oct 2026 00:00:00 +0800 The terms you accept by using this website. https://www.goodailabs.com/blog/four-ways-to-grade-the-same-robot/ https://www.goodailabs.com/blog/four-ways-to-grade-the-same-robot/Fri, 02 Oct 2026 00:00:00 +0800 A placement can reach the goal and fall over afterward. Four historical Samsara protocols tested when success should count. https://www.goodailabs.com/blog/sovereign-intelligence/ https://www.goodailabs.com/blog/sovereign-intelligence/Fri, 02 Oct 2026 00:00:00 +0800 reflex-1 runs on a laptop CPU as well as an H200. The request, latency target and runtime memory determine which machine fits the application. https://www.goodailabs.com/careers/ https://www.goodailabs.com/careers/Fri, 02 Oct 2026 00:00:00 +0800 We're hiring for model research, vision systems and deployments. https://www.goodailabs.com/blog/security-review-is-the-market/ https://www.goodailabs.com/blog/security-review-is-the-market/Fri, 02 Oct 2026 00:00:00 +0800 Our reflex-1 timings start with the model loaded. A deployed service also has startup, queues and tool calls to account for. https://www.goodailabs.com/blog/debugging-by-decomposition/ https://www.goodailabs.com/blog/debugging-by-decomposition/Fri, 02 Oct 2026 00:00:00 +0800 reflex-1's main adaptation run took 75 minutes on one H200. Where the trainable parameters sit and what that timing includes. https://www.goodailabs.com/security/ https://www.goodailabs.com/security/ Model connections, tool permissions and local deployment.