SDK integrations

React

@signet/sdk/react provides SignetProvider, useSignet, useMemorySearch, and useMemory.

import { SignetProvider, useMemorySearch } from "@signet/sdk/react";

function Preferences() {
  const { data, loading, error } = useMemorySearch("user preferences", {
    limit: 5,
    type: "preference",
  });
  if (loading) return <p>Loading…</p>;
  if (error) return <p>{error.message}</p>;
  return <ul>{data?.map((memory) => <li key={memory.id}>{memory.content}</li>)}</ul>;
}

export function App() {
  return <SignetProvider config={{ daemonUrl: "http://localhost:3850" }}><Preferences /></SignetProvider>;
}

useMemorySearch(null) and useMemory(null) suppress their request. SignetProvider checks health() on mount and exposes connected and error through useSignet().

Vercel AI SDK and OpenAI tools

The SDK exports four memory tools: memory_search, memory_store, memory_modify, and memory_forget.

import { SignetClient } from "@signet/sdk";
import { memoryTools } from "@signet/sdk/ai-sdk";
import { generateText } from "ai";

const client = new SignetClient();
const result = await generateText({ model, tools: await memoryTools(client), prompt: "Find relevant preferences." });

For OpenAI function calling, use memoryToolDefinitions() to construct the tool list and executeMemoryTool(client, name, parsedArgs) to dispatch a returned function call. Both adapters use camelCase arguments such as aggregateBudget, sessionKey, and agentId because they call the TypeScript client, not the raw daemon wire schema.

Lifecycle hooks

The SDK uses camelCase request fields. The daemon may accept compatibility wire aliases, but SDK callers should use the exported TypeScript shape.

import { SignetClient } from "@signet/sdk";

const client = new SignetClient();
await client.sessionStart({ sessionKey: "session-123", harness: "my-harness", project: "/workspace/app" });
const context = await client.userPromptSubmit({ sessionKey: "session-123", prompt: "Review the migration", project: "/workspace/app" });
await client.sessionEnd({ sessionKey: "session-123", harness: "my-harness", transcript: "…", project: "/workspace/app" });
await client.preCompaction({ sessionKey: "session-123", context: "Current session summary", project: "/workspace/app" });
await client.compactionComplete({ sessionKey: "session-123", summary: "Compacted summary", project: "/workspace/app" });
await client.requestSynthesis({ project: "/workspace/app" });

sessionEnd requires sessionKey and harness; it does not take a summary field. Synthesis is Dreaming-owned and manifest-gated; callers request a run with requestSynthesis and must not submit generated MEMORY.md content.

Connectors

const created = await client.createConnector({
  provider: "filesystem",
  displayName: "Project notes",
  settings: { rootPath: "/workspace/notes" },
});

await client.syncConnector(created.id);
const health = await client.getConnectorHealth(created.id);

Supported createConnector providers are filesystem, github-docs, and gdrive. Use settings, not retired config, and use getConnectorHealth, not checkConnectorHealth.