{"version":"2.1.282","anchor":"claude-apps-gateway-gains-a-load-test-mode-that-returns-cann","canonical_anchor":"claude-apps-gateway-gains-a-load-test-mode-that-returns-cann","heading":"Claude apps gateway gains a load_test_mode that returns canned replies","tier":"use","area":"Self-Hosted Gateway","url":"https:\/\/changelogs.core-directive.com\/v\/2.1.282\/e\/claude-apps-gateway-gains-a-load-test-mode-that-returns-cann","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.282","markdown":"### Claude apps gateway gains a load_test_mode that returns canned replies\n\nGateway operators can enable load_test_mode, which sends no requests to the model, returns canned replies and can create test users\n\n**What**\n\nThe Claude apps gateway is a server that organisations run between their users and the model providers. Its configuration now accepts an optional `load_test_mode` block. It only takes effect when an operator adds it to the gateway configuration.\n\n- `enabled` turns it on. `reply_tokens` (default 750, maximum 100,000) and `reply_seconds` (default 9.5, maximum 600) set the size and delay of the canned replies.\n\n- While it is on, no model request is sent upstream and every reply is canned. Calls to fetch credentials, such as AWS STS, still go out.\n\n- Callers can send an `x-load-test-user` header holding a whole number of up to 7 digits. The request is then recorded under a made-up identity: the caller's sub plus `+loadtest-N`.\n\n- Inference log lines are tagged `load_test: true`, and spend is not metered.\n\n- The gateway logs a warning at startup when `load_test_mode.enabled` is set.\n\n- At startup it checks that every upstream was built to use the canned replies, and fails to start if one was not.\n\n- It refuses to start load-test mode against a Postgres store (`store.postgres_url`) that already holds real spend.\n\n- The provider clients `anthropicAws`, `vertex` and `foundry` now accept a replacement `fetch` function, which is what makes the canned replies possible.\n\n**Why**\n\nOperators can test how the gateway copes under heavy load without paying for model calls or using real accounts. The startup warning and the checks exist to keep this mode away from a production database and from real users.\n\n- Area: Self-Hosted Gateway\n- Names: `load_test_mode`\n- Tier: Use it now\n- Useful: 4\/5\n- Signal: 3\/5"}