Capture context with provenance
Per-robot state with timestamps, sources and confidence. A stale observation does not become fresh evidence.
Robotics Operational Systems and Agave
Open libraries connecting robot state to commands, agents and local AI. See what happens when the conditions change.
No signup. No hardware. Original code under MIT.
The task requires at least 20% battery. Propose returning to base.
A small, inspectable layer for ROS 2 applications. Command interpretation stays separate from the rules that allow a task to run.
Per-robot state with timestamps, sources and confidence. A stale observation does not become fresh evidence.
Interpret an allowed task. Check capability, battery, obstacles and stop state before requesting confirmation.
If context changes or the proposal expires, request a new one. Retrying does not duplicate simulated effects.
Experimental prototype: the demo simulates tasks, uses rules and does not control motors. It is not a certified safety system.
Names rooted in Guadalajara and its region. Descriptive modules, explicit contracts and open documentation.
context.pyIdentity, fresh observations and SQLite event memory.
Corepolicy.pyStructured intents, deterministic gates and expiring proposals.
Coreplanner.pyNo-LLM rules or interpretation through a local Ollama model.
Adapterros2.pyObserve ROS 2 telemetry and publish context. No actuator commands.
Adaptergateway.pySigned HTTPS context export to your chosen agent platform.
AdapterThe local simulator needs only Python 3.11 or newer. Includes tests, schemas, examples and integration guidance.
Read the documentation →# Clone, inspect, run git clone https://github.com/IgDiaz/ROSA.git cd ROSA python -m rosa # Open http://127.0.0.1:8765
ROSA stands for Robotics Operational Systems and Agave. Agave roots our identity in Jalisco; CDT shares this toolkit to build, learn and collaborate in the open.
ROS 2, Ollama and external endpoints need validation in your environment. Read the limitations and share reproducible findings.