#how to make custom tools be called fast?

1 messages · Page 1 of 1 (latest)

grand lynxBOT
#
AI-generated Answer for jvini6832

Short answer: make the model decide instantly, avoid human/tool blocking, and keep the webhook path fast. Concretely:

  • Remove approval friction for trusted tools. If you integrate via MCP, set approval to No Approval or mark individual tools Auto-approved so the agent can execute without waiting for permission. Use Fine-Grained Tool Approval to keep only risky tools gated. (elevenlabs.io)
  • Don’t block on client tools unless you must. Only enable “Wait for response” on a client tool when the agent truly needs the return value; otherwise leave it off so the conversation continues while your code runs. (elevenlabs.io)
  • Set a strict webhook timeout. When creating server/webhook tools, configure response_timeout_secs so the agent fails fast instead of waiting on slow upstreams; you can monitor avg_latency_secs for each tool to verify improvements. (elevenlabs.io)
  • Help the LLM pick the right tool immediately. Use clear tool names/descriptions, well-described parameters, and put calling rules (“use X when…, call Y before Z…”) in the agent’s system prompt. This reduces indecision and extra turns before the call. (elevenlabs.io)
  • Choose an LLM optimized for tool calling and low latency. Docs recommend high‑intelligence models for reliable function calls; ElevenLabs’ platform blog currently recommends Gemini 2.5 Flash as the default for low latency and robust tool calling. (elevenlabs.io)
  • Use async patterns for long tasks. Have the agent acknowledge the action (“Let me check that…”) and complete it via webhook asynchronously rather than blocking the turn; this keeps the interaction feeling instant while the tool finishes. (elevenlabs.io)
  • If exposing tools via MCP, prefer SSE/HTTP streamable transports to support real‑time interactions. (elevenlabs.io)

If you share which tool type (Client, Webhook/Server, or MCP) you’re using and your current latencies, I can point to the exact settings and API payloads to adjust. (elevenlabs.io)

Sources:
Model Context Protocol | ElevenLabs Documentation
Client tools | ElevenLabs Documentation
Create tool | ElevenLabs Documentation
Server tools | ElevenLabs Documentation
Server tools | ElevenLabs Documentation
How do you optimize latency for Conversational AI? | ElevenLabs
Model Context Protocol | ElevenLabs Documentation
Tools | ElevenLabs Documentation

craggy pelican
#

how many documents a single agent can use from knowledgebase