Local-first when it matters
Core operations are not automatically forced to depend on connectivity or a cloud vendor.
Lab manifesto / BirruLabs
Not a landing-page factory. A lab for understanding how AI should work.
BirruLabs is an independent product lab in Indonesia. We explore agent systems, operational automation, and local-first software with one standard: people must remain able to understand and control the system.
01 / Engineering principles
People own critical decisions
Agents may propose and execute within bounds; risky approvals remain explicit.
Behavior must be observable
Inputs, tool use, failures, and outputs must leave a trail operators can review.
Contracts over prompts
Schemas, permissions, validation, and failure modes are product decisions—not implementation trivia.
Least access by default
Workers receive only the context and tools required for their task.
Stages without spin
Research, Prototype, and Internal Alpha mean what they say. Stage is not decoration.
02 / Lab thesis
The AI advantage is not the model that looks smartest. It is the system that remains operable when reality gets messy.
That is why we design permissions, observability, retries, provenance, and human gates alongside the user experience—not after the demo works.
See the systems in progress