README

Sprix SAGE Router

 Checkpoint-aware mid-execution rerouting for open A2A networks






An open-source research output of Sprix AI at 屿智同行.

Choose whether an in-flight task should continue, recruit collaborators, or hand off after accounting for completed DAG nodes, reusable artifacts, observed partial quality, remaining work, failures, budget, and deadline.

Quick start · Algorithm · Related work · A2A integration · Operations · Benchmark · Changelog · Contributing



---

 Why SAGE?

Agent discovery tells a system which agents exist. It does not answer the harder runtime question: who should work with whom after execution has already begun?

SAGE—State-Aware Graph Exchange—is the decision layer between A2A discovery and task execution. It evaluates three routes in one auditable objective:

| Route | Ownership | Best used when |
|---|---|---|
| SELF | Incumbent agent | Existing capability and accumulated context are sufficient |
| COLLABORATE | Incumbent retains ownership | A small complementary team covers missing requirements |
| HANDOFF | A peer takes full ownership | Specialist advantage exceeds context-transfer loss |

SAGE is designed to sit above the Agent2Agent (A2A) protocol. A2A provides Agent Cards, messages, tasks, artifacts, authentication, and transport. SAGE decides which feasible agent configuration should execute the task, in which mode, and why.

!SAGE routing pipeline and evidence loop

Figure 1. SAGE filters candidates, compares all three routing modes, jointly searches assignments and schedules, ranks feasible plans, and learns from execution evidence.

 Research focus

SAGE is deliberately narrow: checkpoint-aware reconfiguration after execution has begun.

- Concrete continuation value. The router preserves completed requirements and combines in-flight completion, observed partial quality, current ownership, and artifact portability to estimate how much work each candidate must actually redo.
- Comparable runtime actions. SELF, COLLABORATE, and HANDOFF share one permission-, budget-, and deadline-constrained action space. Progress-masked and static-coalition baselines receive the same registry and limits.
- Requirement-conditioned evidence. Reliability is tracked per agent and requirement instead of assuming that one reputation score transfers across skills.
- Trajectory-level falsification. A separate evaluator replays checkpoints and scores artifact reuse, added cost, recovery latency, wasted work, and final quality without calling SAGE's switching equation.

Noisy-OR, beam search, a linear utility, Beta beliefs, online logistic regression, and DAG-induced communication edges are not claimed as inventions. They are replaceable implementation mechanisms. The repository also provides permission-first filtering, bounded candidate search, workload-sensitive quotes, auditable alternatives, degraded-route flags, state persistence, and transport-neutral A2A plans as engineering features.

 Core algorithm

For task requirement

Top issues

Latest release

Benchmark rigor and scalability

- structurally held-out geometric evaluator with independently specified latent skills
- random and greedy team baselines
- learning, exploration, initialization, and early/late ablations
- bounded top-k candidate prefilter and request-local scoring caches
- workload-sensitive cost and explicit authorized degraded-route flags
- evidence-gated pair synergy updates
- related-work comparisons, module split, ruff/mypy CI, and release automation

The benchmark remains an author-designed synthetic suite, not evidence of real-world superiority. PyPI upload stays gated