Separate direction from execution
Gravis provides the sprint loop. DevBridge provides an execution runtime with CLI, API and MCP interfaces. Keeping those responsibilities separate makes each boundary easier to reason about.
Build. Verify.
Recover. Repeat.
An AI development ecosystem separating the orchestration of work from the runtime that executes it.
GRAVIS / EXECUTION TRACEILLUSTRATIVE RUN
~/gravis / demonstration
> Awaiting a scoped task. > Select “Run the trace” to follow the handoffs.
No model calls. No files changed.
Gravis + DevBridge
An agent producing code is only one part of a useful development workflow. The surrounding system has to scope the task, expose progress, validate changes and decide what happens when a check fails.
Bound the recovery loop
DevBridge includes a consecutive-failure budget. It records a root-cause hint, stops when the budget is exhausted and resets the counter after a successful verification.
Make the loop observable
The Gravis entrypoint connects routing, budget, event-log and validation-readiness modules. The architecture makes room for an operator to inspect and steer work.
01Task & constraints
02Execution runtime
03Validation
04Recovery / report
Follow the execution loop.
Step through a successful check or a bounded recovery path. This is an explainer, not a live AI run.
Current scope
This is independent tooling in development. The interactive sequence below is an architecture illustration, not a live agent session or benchmark.
Evidence reviewed October 2026. Source inspection is not a substitute for runtime validation.