Briarios
Security-native orchestration meta-layer for parallel agent workstreams
triage scoring, model-tier routing, budget gating, and independent verification
What it is
Briarios is a security-native orchestration meta-layer for parallel agentic development workstreams.
It does not replace OpenHands, LangGraph, or managed agent platforms. It sits above them and adds the operational controls needed to run several agent lanes at once: security-aware triage, model-tier routing, multi-resource budget gates, and independent verification before human approval.
What it adds
Security-aware triage
Issues are scored by security risk, blast radius, complexity, and blocking impact before assignment.
Model-tier routing
High-risk and ambiguous work routes to stronger models; small, bounded tasks can use cheaper lanes.
Budget gates
Parallel agent lanes are gated by token budget, CPU load, VRAM headroom, and external API limits.
Verification pipeline
Diffs pass through test, review, eval, and security checks before crossing a human merge gate.
Triage model
Every issue is scored across four axes:
| Axis | What it captures |
|---|---|
| Security risk | Cosmetic change vs. exploitable vulnerability |
| Blast radius | Isolated file vs. cross-cutting subsystem |
| Complexity | Clear implementation vs. architecture decision |
| Blocking | Standalone task vs. blocker for other work |
The score controls agent tier, verification depth, and whether a security-focused review lane is required.
Status
Research complete. Architectural decisions are accepted and the project is ready for scaffolding.
See the GitHub repo for the current ADRs and research notes.