{"id":"agience-agience-core","name":"agience-core","af_score":34.8,"security_score":61.2,"reliability_score":20.0,"what_it_does":"Agience Core is an open, extensible runtime for agentic AI centered on shared, typed, versioned “artifacts” with provenance and human-in-the-loop workflows. It is positioned as both an MCP server (exposing tools) and an MCP client (consuming vendor MCP servers), and emphasizes capability-based access via scoped API keys and delegations tracked by authority/provenance.","best_when":"You need an auditable shared-state runtime where agents and humans operate on the same typed artifacts, with MCP-based extensibility and governance/provenance modeled as first-class architecture.","avoid_when":"You need a lightweight, minimal integration without standing up an MCP server/runtime, or you cannot accept AGPL obligations for network deployment/modification.","last_evaluated":"2026-03-30T15:27:06.322531+00:00","has_mcp":true,"has_api":false,"auth_methods":["Scoped API keys (capability-based access; as described in README)","Identity derived from auth token (as described in README)","Delegated operations with record of who authorized them (as described in README)"],"has_free_tier":false,"known_gotchas":["Provided README does not include concrete MCP server tool schemas, transport details, or error-handling conventions; agents may need to infer behaviors until full docs are available.","Idempotency/retry guidance is not documented in the provided content; agent runtimes should implement cautious retry logic for any state-changing operations.","Because it is positioned as a runtime with artifacts/provenance and human approval gates, agent flows may require explicit commit/approval steps to avoid unexpected “denied until commit” behavior."],"error_quality":0.0}