TencentCloud/TencentDB-Agent-Memory · error · LlmResolveError

llm.provider=proxy 需要 instanceId，但 core 当前 instanceId 为空 ——

Error message

llm.provider=proxy 需要 instanceId，但 core 当前 instanceId 为空 —— service 模式下确保请求带 x-tdai-service-id，standalone 模式下确保 yaml 有 instanceId

What it means

With llm.provider=proxy, the gateway needs an instanceId to build the per-instance proxy path /proxy/<iid>/v1. When the current instanceId is empty — i.e. no x-tdai-service-id header in service mode and no instanceId in standalone yaml — resolveEffectiveLlmConfig throws this LlmResolveError before any LLM call.

Source

Thrown at MemoryCore/src/gateway/llm-resolver.ts:60

export function resolveEffectiveLlmConfig(
  llm: StandaloneLLMConfig,
  instanceId: string | undefined,
  memorySystemUser: MemorySystemUserConfig | undefined,
): StandaloneLLMConfig {
  const provider = llm.provider ?? "openai";
  if (provider !== "proxy") {
    // 默认路径：原样返回，行为完全等价改造前
    return llm;
  }

  // provider=proxy 的校验
  if (!llm.baseUrl) {
    throw new LlmResolveError(
      "llm.provider=proxy 需要 llm.baseUrl 指向 context_proxy 根 URL (如 http://127.0.0.1:8096)",
    );
  }
  if (!instanceId || !instanceId.trim()) {
    throw new LlmResolveError(
      "llm.provider=proxy 需要 instanceId，但 core 当前 instanceId 为空 —— " +
      "service 模式下确保请求带 x-tdai-service-id，standalone 模式下确保 yaml 有 instanceId",
    );
  }

  const useSystemUserKey = llm.proxy?.useMemorySystemUserKey ?? true;
  let effectiveApiKey = llm.apiKey;
  if (useSystemUserKey) {
    if (!memorySystemUser) {
      throw new LlmResolveError(
        "llm.provider=proxy 且 llm.proxy.useMemorySystemUserKey=true 需要 " +
        "metadata.systemUser.memory 完整配置（userId + userKey），当前缺失",
      );
    }
    if (!isValidMemorySystemUserKey(memorySystemUser.userKey)) {
      throw new LlmResolveError(
        "metadata.systemUser.memory.userKey 必须匹配 sk-mem-[A-Za-z0-9_-]{32}",
      );

View on GitHub (pinned to 3efcd317b8)

Solutions

1. In service mode, send the x-tdai-service-id header on every request to the gateway (with the registered service instance id).
2. In standalone mode, set instanceId in the gateway yaml so the core has a non-empty id.
3. Check for ingress/load-balancer rules that strip x-tdai-service-id before it reaches MemoryCore.
4. Verify which config file is actually loaded (startup logs) — instanceId may be set in a different yaml than the active one.

Example fix

// before (standalone yaml)
llm:
  provider: proxy
  baseUrl: http://127.0.0.1:8096
// after
instanceId: my-core-instance
llm:
  provider: proxy
  baseUrl: http://127.0.0.1:8096
// service mode: curl -H "x-tdai-service-id: my-core-instance" ...

const iid = process.env.INSTANCE_ID ?? yamlConfig.instanceId;
if (llmConfig.provider === 'proxy' && !(iid && iid.trim())) {
  throw new Error('provider=proxy requires a non-empty instanceId (yaml instanceId or x-tdai-service-id header)');
}

function hasInstanceId(v: unknown): v is string {
  return typeof v === 'string' && v.trim().length > 0;
}

try {
  await gateway.start();
} catch (err) {
  if (err instanceof LlmResolveError && err.message.includes('instanceId')) {
    console.error('Set yaml instanceId (standalone) or ensure x-tdai-service-id header (service mode)');
  }
  throw err;
}

When it happens

Trigger: provider=proxy with a valid baseUrl, but instanceId is empty: service-mode requests missing the x-tdai-service-id header, or standalone deployments whose yaml has no instanceId field (or it is whitespace).

Common situations: Calling the gateway directly via curl/SDK without the x-tdai-service-id header; new standalone deployment where the instanceId section was never filled; header stripped by an ingress/proxy; instanceId set only in a different config file than the one loaded.

Related errors

• llm.provider=proxy 需要 llm.baseUrl 指向 context_proxy 根 URL (如
• llm.provider=proxy 且 llm.proxy.useMemorySystemUserKey=true 需
• llm.provider=proxy 且 useMemorySystemUserKey=false 时必须显式配置 ll
• llm.provider=proxy 需要 llm.baseUrl 指向 context_proxy 根 URL
• llm.provider=proxy 且 useMemorySystemUserKey=true 需要 metadata

AI-assisted analysis of TencentCloud/TencentDB-Agent-Memory@3efcd317b8 (2026-09-01). Data as JSON: /api/errors/5351c37064ebbcf9. Report an issue: GitHub.