Sprix AI Releases A2A Router

Sprix AI has released Sprix SAGE Router, a Python library that determines the optimal action path for agents running in an A2A (Agent-to-Agent) protocol environment. SAGE (State-Aware Graph Exchange) compares three paths—SELF (execute alone), COLLABORATE (collaborate), and HANDOFF (hand off)—against a single objective function, considering factors such as current task progress, failure history, and context loss costs, to establish an optimal execution plan.

Key Features and Design

• State-Aware Routing: Instead of simple capability assessment, it uses completed task nodes and accumulated context as decision inputs. Notably, it calculates 'context transfer loss' incurred during HANDOFF as an explicit cost term.
• Differentiated Team Composition Algorithm: Rather than gathering agents with high individual rankings, it rewards the degree to which requirements are newly covered (marginal coverage). Reliability is learned separately per agent and per requirement, ensuring that success in one domain does not guarantee success in another.
• Pipeline Structure: Execution proceeds in the following order: eligibility gate (checking permissions, budget, and deadlines) -> candidate set generation -> role assignment and schedule search -> final ranking based on the objective function. Agents that fail to pass hard filters are excluded regardless of their scores.

Usage and Characteristics

• No Runtime Dependencies: After git clone, it can be run immediately on Python 3.10 or higher.
• Audit and Learning Support: The route_with_trace function records not only the selected path but also rejection reasons and alternative rankings. It updates reliability and estimation accuracy by reflecting execution results (record_outcome), and the learning state is saved as a JSON snapshot.
• Research Preview Status: Currently an early-stage research artifact, actual protocol transmission and authentication must be handled on the A2A client side.