# Journal · Good AI Labs

> Model announcements, product updates and notes from our research.

URL: https://www.goodailabs.com/blog/

## Pages

- [Introducing Reflex-1](https://www.goodailabs.com/blog/reflex-1-fast-decisions/): A 421M-parameter model for tool routing, intent classification and action selection, with public weights and measured CPU performance.
- [Introducing Samsara-VLA](https://www.goodailabs.com/blog/samsara-capability-and-compute/): Robot control with memory, in two compact models.
- [Introducing re1: spatial memory for physical AI](https://www.goodailabs.com/blog/re1-video-memory/): Search recorded events, check the source video and revisit captured places in 3D.
- [Selecting Samsara's checkpoint](https://www.goodailabs.com/blog/a-number-you-cannot-tune-for/): 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.
- [Introducing Colony: personal agents with memory and tools](https://www.goodailabs.com/blog/colony-runtime-and-model-budget/): Give an agent an ongoing job, a place to work and memory it can use next time. Colony is opening for technical previews.
- [Adapting 9M parameters inside a 421M model](https://www.goodailabs.com/blog/debugging-by-decomposition/): reflex-1's main adaptation run took 75 minutes on one H200. Where the trainable parameters sit and what that timing includes.
- [Budgeting a local decision service](https://www.goodailabs.com/blog/security-review-is-the-market/): Our reflex-1 timings start with the model loaded. A deployed service also has startup, queues and tool calls to account for.
- [Choosing hardware for a local model](https://www.goodailabs.com/blog/sovereign-intelligence/): 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.
- [Four tests for a robot placement](https://www.goodailabs.com/blog/four-ways-to-grade-the-same-robot/): A placement can reach the goal and fall over afterward. Four historical Samsara protocols tested when success should count.
- [Why we build local models](https://www.goodailabs.com/blog/goodness-first/): 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.

Contact: research@goodailabs.com