2 min read
Writing from the lab
Journal
Model announcements, product updates and notes from our research.
ANNOUNCEMENTS & LAB NOTESSubscribe via RSS
Reflex-1 · Original Chromium ↗Announcement Decision models
Introducing Reflex-1
A 421M-parameter model for tool routing, intent classification and action selection, with public weights and measured CPU performance.
5 min read
Read the announcement Announcements
3 ARTICLES
Spatial memory
Introducing re1: spatial memory for physical AI
Search recorded events, check the source video and revisit captured places in 3D.
3 min read
Agent systems
Introducing Colony: personal agents with memory and tools
Give an agent an ongoing job, a place to work and memory it can use next time. Colony is opening for technical previews.
3 min read
Lab notes
6 ARTICLESEvaluation
Selecting Samsara's checkpoint
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.
2 min read
The lab
Why we build local models
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.
1 min read
Robot learning
Four tests for a robot placement
A placement can reach the goal and fall over afterward. Four historical Samsara protocols tested when success should count.
2 min read
Systems
Choosing hardware for a local model
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.
2 min read
Systems
Budgeting a local decision service
Our reflex-1 timings start with the model loaded. A deployed service also has startup, queues and tool calls to account for.
2 min read
Model engineering
Adapting 9M parameters inside a 421M model
reflex-1's main adaptation run took 75 minutes on one H200. Where the trainable parameters sit and what that timing includes.
1 min read