Support & Escalation

Run through this triage checklist first. Support will ask you these questions anyway — coming prepared saves a full round-trip:

1. Check GET /health — is the platform operational? If any component is "unhealthy", the issue may resolve itself once infrastructure recovers.
2. Read the error code — look at error.code in the response body. Is it VALIDATION_ERROR, RATE_LIMIT_EXCEEDED, TENANT_CONTEXT_REQUIRED? Each code has specific fixes documented in Common Errors.
3. Copy the X-Request-ID — from the response headers of the failing request. This is the single most valuable piece of information for the support team.
4. Check the audit log — in Dashboard → Audit Logs, filter by your resource type and look for entries with success: false around the time of the failure.
5. Check your billing status — in Dashboard → Billing. If you have hit your monthly quota, you may be in silent mode — new add requests return 200 OK but no memory is stored or embedded, and queries return empty results. This is the #1 cause of “my queries return no results” reports.
6. Confirm the environment — are you using the same environment (development/production) and project for both writes and reads?

When you do contact support, include all of the following. Missing information means the support team has to ask for it, adding a full reply cycle to the resolution time:

# Good support request template:
Request ID:   a1b2c3d4-e5f6-7890-abcd-ef1234567890
Trace ID:     trace-xyz-123
Timestamp:    2026-05-09T14:30:00Z (UTC)
Endpoint:     POST /memory/query
Environment:  production
Project:      proj_abc123

Error response:
{
  "error": { "code": "...", "message": "..." },
  "request_id": "a1b2c3d4-..."
}

Request body (redacted):
{ "query": "user preferences", "user_id": "user_42" }

The X-Request-ID and X-Trace-ID headers form the backbone of MemorySync's observability. Here's how support uses them to trace your issue:

Single-request issues (e.g., "this specific call failed"):

• Support searches by X-Request-ID to find the exact request in server logs
• The request log includes: method, path, status code, duration, user ID, tenant ID, and any errors
• If the request triggered background work (such as a webhook dispatch), those operations are linked by the same trace ID.

Cross-system issues (e.g., "I stored a memory but the webhook didn't fire"):

• Support searches by X-Trace-ID to find all related operations
• This reveals the full chain: API request → memory stored → webhook dispatched → delivery attempted.
• The exact point of failure in the chain becomes visible

The audit log is your first line of evidence before contacting support. Every mutating API call is recorded with a structured schema that makes it searchable:

Useful audit log queries for debugging:

Sharing relevant audit log entries with support (screenshot or export) gives them immediate context without needing to search server logs.

Support response times and channels depend on your plan tier:

Severity classification:

• Critical — all API requests failing, data loss suspected, security incident. Triggers immediate escalation on Priority/Dedicated tiers.
• High — a specific feature is broken for your organization (e.g., all webhook deliveries failing)
• Medium — intermittent errors, performance degradation, non-blocking issues
• Low — questions, feature requests, documentation issues

These are the most frequent "false alarms" that reach the support team. Check for these before escalating:

For security and data integrity reasons, the support team has the following limitations:

• Reset billing counters mid-cycle — usage counters are atomic and tamper-proof. They reset automatically at the start of each billing cycle. If you've hit your limit, upgrade your plan or wait for the next cycle.
• Bypass rate limits — rate limits are enforced at the infrastructure level and cannot be selectively disabled. Upgrade your API key's tier or distribute load across multiple keys.
• Access or return raw memory content — support can trace request flow and identify errors, but cannot read or export your stored memory data. All data access must go through the API with proper authentication.
• Override tenant isolation — each tenant's data is strictly isolated. Support cannot move memories between tenants, merge tenant data, or grant cross-tenant access.
• Modify the ranking algorithm per-tenant — the ranking weights, importance floor, and semantic dedup threshold are system-wide constants. They cannot be customized for individual organizations.