AI-generated Answer for bibek_chaulagain_82978
Thanks for the details. Continuous tool-call failures in ElevenLabs Agents typically come down to one of these documented issues:
-
Hitting concurrency limits (429 errors). Check the Call History for the exact failure reason and look for 429 too_many_concurrent_requests or system_busy. Agents have per‑workspace concurrency caps by plan; you can raise headroom with Burst Pricing or upgrade concurrency. See Call History, Agents concurrency limits, 429 error codes, and Burst Pricing. (help.elevenlabs.io)
-
Tool timeouts. Tools have a configurable response_timeout_secs; client‑tool response timeout was increased to 120s, and webhook/server tools expose response_timeout_secs in the tool config. If your tool is invoked back‑to‑back, make sure each call returns within this window or increase the timeout. (elevenlabs.io)
-
Webhook behavior (server tools). Your webhook should return HTTP 200 quickly; repeated failures can auto‑disable the webhook until you re‑enable it in settings. If calls are failing continuously, check whether the webhook has been auto‑disabled. (elevenlabs.io)
-
LLM choice affecting function/tool reliability. For tool calling, ElevenLabs recommends high‑intelligence models (e.g., GPT‑4o mini, Claude 3.5 Sonnet) and specifically advises avoiding Gemini 1.5 Flash. If you’re seeing aggressive/looping tool calls or poor argument extraction with Gemini Flash, try one of the recommended models to compare behavior. (elevenlabs.io)
What to do next (fast checks):
- Open Call History and copy the exact error shown for the failed turns. (help.elevenlabs.io)
- If it’s 429/system_busy, reduce parallel tool invocations (debounce/retry with jitter) or enable Burst Pricing on that agent. (help.elevenlabs.io)
- If it’s a timeout, raise response_timeout_secs on the tool and ensure your endpoint responds within that SLA. (elevenlabs.io)
- If it’s repeated webhook failure, re‑enable the webhook and make sure you return 2xx promptly. (elevenlabs.io)
- If tool selection/arguments look wrong, try a recommended LLM for tool use. (elevenlabs.io)
little wraith
latent pilot