Platform & SDK
SDK
The private, shared package layer every Runline service, Runner, and integration builds on.
The Runline SDK (Runline-AI/sdk) is the private, shared package layer every Runline service, Runner, and integration builds on: an npm-workspaces monorepo of independently versioned @runline-ai/* packages.
Per ADR-006, Runline's code splits into three layers: the Platform (Arc), the SDK (these shared packages), and per-institution Deployments. The SDK is Layer 2, the shared toolkit both our own agents and a client's Runners consume.
Packages
| Package | Purpose |
|---|---|
@runline-ai/attention-queue | Priority-scored attention queue: what a Runner works on next. |
@runline-ai/agent-memory | Hybrid search, learnings engine, context builder, synthesis. |
@runline-ai/connections | The client for the brokered Connections layer. |
@runline-ai/domain-taxonomy | The canonical CU domain and Runline slugs. |
More land as the patterns stabilize: shared agent config, compliance validators, core-processor connectors.
Design principles
- Independently versioned and published. A new
agent-memoryrelease doesn't require a platform deploy. - Optional peer dependencies. Install only the engines you use (DuckDB, LanceDB, Voyage).
- Dual ESM/CJS, so any consumer can import it.
- CLI-first, API-second. Every operation works from the terminal; we prefer thin CLI wrappers over MCP.
Reference
- ADR-006 · Three-Layer Repository Architecture
- Orientation & Repo Map