@innerlife/channel-engine-adapter

Implements ChannelEngine for an inbox + runner brain (Innerlife Agent). Bridges @innerlife/channel's host to the brain: inbound → inbox.push → per-agent serial runner.run → outbound. Zero dependency on @innerlife/agent — it talks to a minimal structural BrainAgent / BrainRunner, so any matching brain works and the app supplies the concrete instances.

Behaviour (see DESIGN.md §7)

• Identity — resolveAgent(address, msg) → { agent, source } decides who.
• Merge — push straight into the brain's inbox (mergeKey = chat:<source>); the inbox merges bursts natively. A small debounceMs collects the first burst.
• Concurrency — serial per agent instance (runner.run isn't reentrant), concurrent across agents, bounded by a global semaphore (concurrency, default 8).
• Media — mediaToText(ref) downgrades attachments to text (inbox text is a string).
• Proactive — engine.notify({ agent, source, text }) (path B).

Reply routing assumes one agent per conversation source (the recommended model). A shared agent serving many sources cannot route replies precisely until the brain surfaces the processed source.

Usage

import { createInnerlifeEngine } from "@innerlife/channel-engine-adapter";
import { ChannelHost } from "@innerlife/channel";
import { weixin } from "@innerlife/channel-weixin";

const engine = createInnerlifeEngine({
  runner,                                   // your Innerlife Runner
  resolveAgent: async (addr) => ({ agent: await pool.get(addr.peerId), source: addr.peerId }),
  // mediaToText, resolveAddress, concurrency, debounceMs ... optional
});

const host = new ChannelHost({ channels: [weixin()], engine });
await host.start();