nla-adapters

Portable adapter packages for NLA.

The split in this repo is:

• @nla/sdk-core generic NLA adapter/runtime mechanics
• @nla-adapters/contracts neutral capability contracts
• @nla-adapters/todo-agent portable todo adapter logic that depends only on NLA plus contracts

host-runtime should not be the portable surface. It should be one capability implementation behind thin wrappers.

Current Packages

packages/
  contracts/   neutral capability contracts
  todo-agent/  portable todo adapter
  wallet-agent/ portable wallet adapter

Todo Agent Shape

The todo adapter is created by dependency injection:

import { createTodoAgent } from "@nla-adapters/todo-agent";

const adapter = createTodoAgent({
  createModel: () => myToolLoopModel(),
  storage: {
    getJson: async (request) => { ... },
    putJson: async (request) => { ... }
  }
});

For stdio processes:

import { runTodoAgentStdio } from "@nla-adapters/todo-agent/stdio";

await runTodoAgentStdio({
  createModel: () => myToolLoopModel(),
  storage: {
    getJson: async (request) => { ... },
    putJson: async (request) => { ... }
  }
});

Host Runtime Wrapper Example

This is the intended wrapper style inside host-runtime:

import * as Effect from "effect/Effect";
import { runTodoAgentStdio } from "@nla-adapters/todo-agent/stdio";
import { toolLoopModel } from "@host-runtime/sdk/llm";
import { getJson, putJson } from "@host-runtime/sdk/storage";

await runTodoAgentStdio({
  createModel: () =>
    toolLoopModel({
      provider: "openrouter",
      metadata: {
        agent_id: "todo.agent"
      }
    }),
  storage: {
    getJson: (request) => Effect.runPromise(getJson(request)),
    putJson: (request) => Effect.runPromise(putJson(request))
  }
});

That keeps the portable adapter free of @host-runtime/sdk/* while still letting host-runtime provide storage and model brokering.