Project overview and objectives
Kyostra is building an execution governance layer for robots and humanoid robots.
Our first product, Kyostra Invar, runs at the actuator level. It sits below planners and AI, above actuator drivers, and governs how motion is physically executed. The core thesis is straightforward: in robotics, trust does not stop at decision-making. It also depends on whether physical execution remains bounded when hardware faults appear, communication degrades, or upstream behavior becomes unstable.
For 0G, the relevant part of our work is the verifiability layer around execution.
We want to make robot execution attestable across three stages:
-
During fault-injection testing, record the exact test conditions, the interventions triggered, and the resulting outputs in a structured and replayable way
-
In deployed robots, provide a black-box style event record for audit, liability analysis, and post-incident reconstruction
-
Extend this toward AEGIS, our planned fleet-level auditability and liability layer, where multiple robots, operators, and counterparties may need a neutral record of what happened
The objective is not to put robot control onchain. The objective is to create cryptographically verifiable evidence of physical execution.
Technical architecture and implementation plan
Kyostra Invar runs locally on actuator-class microcontrollers. It validates commands, enforces authority envelopes, manages local reflexes, tracks energy, and records execution-relevant events. This runtime must remain deterministic and local. It is not something we would push into a blockchain environment.
The blockchain-linked layer starts after execution.
Each node produces signed execution events. These events describe the command context, the active motion constraints, the enforcement actions taken locally, and the fault or reflex transitions that occurred. During validation, those traces are generated through adversarial and fault-injection campaigns. In deployment, they become the basis for audit and incident reconstruction.
Our implementation plan is:
-
Extend simulation and hardware fault-injection workflows so they generate structured execution traces
-
Add signing, batching, and integrity commitments around those traces
-
Anchor those commitments through 0G
-
Demonstrate replay and verification of a robot action against the execution constraints active at the time
This preserves the real-time requirements of embedded control while making the resulting evidence independently verifiable.
How we’ll integrate with 0G infrastructure
We would use 0G as a neutral attestation and anchoring layer for robot execution records.
0G would not sit inside the control loop. Invar keeps running locally on the robot side, with deterministic timing. The interaction with 0G happens at the trace and proof layer.
Concretely, we plan to use 0G for:
-
anchoring signed execution log commitments
-
timestamping and external integrity verification
-
making audit records shareable across parties without relying only on one vendor-controlled backend
This is useful in three contexts.
First, validation. Fault-injection campaigns become structured evidence of bounded execution rather than internal test artifacts.
Second, deployment. Execution records can show what was commanded, what was authorized, what was clipped, what reflex or fallback was triggered, and when.
Third, future fleet operations. Once robots operate across organizational boundaries, a neutral integrity layer becomes more important than a vendor-local log database.
Team background and experience
Jean-Charles Cabelguen is the founder of Kyostra and leads the execution governance architecture. His background spans deeptech adoption, distributed systems, and trusted computing. Particularly relevant here are his experience at iExec and his leadership work within the Enterprise Ethereum Alliance on off-chain trusted compute. That background is directly aligned with the problem of verifiable execution, integrity, and cross-party trust.
Nicholas Witham leads hardware integration. His background in biomechatronics, embedded systems, and actuators is important because the attestation layer has to stay grounded in real actuator behavior and embedded constraints, not only software abstractions.
Together, the team combines execution control, embedded actuation, distributed systems, and trusted infrastructure.
Funding requirements and milestones
We are seeking support to build the first execution attestation pipeline around Kyostra Invar.
The main milestones are:
-
structured execution traces from simulation and fault-injection testing
-
signed event pipeline tied to the runtime
-
first 0G anchoring and verification workflow
-
demonstration that a robot action can be replayed and verified against the governing constraints active during execution
-
preparation for the extension toward AEGIS and fleet-level auditability
The funding is tied to a specific outcome: making physical execution verifiable, first in testing, then in deployed robots, and later across fleets.