> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.talqora.com/operate-at-scale/reliability-and-idempotency/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.talqora.com/_mcp/server.

# Reliability and idempotency

Network retries are normal. Every write and delete requires an `Idempotency-Key` header with at least eight characters.

Talqora stores the request fingerprint and response. Retrying the same request with the same key returns the original response without writing a vector twice or incrementing usage twice.

```bash
curl https://api.talqora.com/v1/indexes/idx_example/vectors \
  -X POST \
  -H "Authorization: Bearer $TALQORA_API_KEY" \
  -H "Idempotency-Key: batch-2026-08-09-01" \
  -H "Content-Type: application/json" \
  -d @batch.json
```

If you receive `429`, honor `Retry-After` before retrying. Do not retry immediately in a tight loop. For network failures, retry with exponential backoff and reuse the same idempotency key.

Every rate-limited response includes the active budget:

```http
RateLimit-Limit: 60
RateLimit-Remaining: 0
RateLimit-Reset: 27
Retry-After: 27
```

`RateLimit-Reset` and `Retry-After` are seconds until capacity becomes available. Successful requests include the first three headers, so clients can slow down before receiving a `429`.