Back to directory

Sprix SAGE Router

An open-source research decision layer for agent-to-agent networks that decides, mid-task, whether an in-flight piece of work should continue with the current agent, recruit collaborators, or hand off to a specialist.

Sprix SAGE Router

Sprix SAGE Router sits above the Agent2Agent (A2A) protocol, which handles agent discovery, messages, tasks, artifacts, authentication and transport, and answers a question A2A discovery alone does not: which feasible combination of agents should execute a task, in what mode, and why. It evaluates three routes in one auditable objective: SELF, where the incumbent agent continues with its existing capability and context; COLLABORATE, where the incumbent keeps ownership but recruits a small complementary team; and HANDOFF, where a peer agent takes full ownership because its specialist advantage outweighs the cost of transferring context.

The router preserves completed requirements and combines in-flight completion, observed partial quality, current ownership and how portable existing artifacts are to estimate how much work a candidate agent would need to redo, tracks reliability per agent and requirement rather than a single reputation score, and is evaluated by a separate trajectory-level evaluator that replays checkpoints against baseline routing strategies under the same registry and limits.

More in Agent frameworks

See all 902
  • Recurse

    recurse.run

    A serverless harness that lets a coding agent design, test, and deploy its own specialist sub-agents as tools, MCP servers, or bots, defined by a schema-validated manifest.

  • ProofForge

    shulgin.is-a.dev

    An AI-agent pipeline that decomposes a mathematics problem, proves each piece, and formalizes the result in Lean 4 and Mathlib until the Lean kernel accepts it, with six of its proofs already merged into Google DeepMind's formal-conjectures repository.

  • local-coder

    github.com

    Builds a local coding-agent team on top of Ollama and OpenCode by detecting the machine's hardware, recommending models for seven specialist roles, and testing whether each assigned model can actually use tools, edit a repository and pass verification.