Thanks — I understand you want to avoid the two-step “call tool → wait for result → generate reply” flow and instead have the agent optimistically assume a write succeeded so it can speak immediately. Short answer: LiveKit Agents do not support bypassing the tool-result processing step. The agent’s ReAct-style loop always executes the tool, returns the tool’s result to the LLM, and then generates the agent response. There is no built-in “fire-and-forget” or optimistic mode that skips that confirmation. Workarounds you can use: • Use an immediate filler reply while the tool runs (for example, call

session.say("Working on that now…")

or

session.generate_reply()

), then after the tool finishes send the final confirmation. This reduces perceived latency while preserving the execution-confirmation loop. • Make the tool return quickly by doing the write asynchronously on the backend (enqueue a background job) and return a fast acknowledgement to the agent. The agent still receives a result, but the write happens off the critical path. • If you don’t need the LLM to know the tool output at all, have the tool return a minimal acknowledgement (or

None

) and mark results silent where applicable; the LLM still processes the acknowledgement but won’t get detailed data. If you’d like, tell me which of the above fits your architecture (filler speech, background job enqueue, or silent minimal acknowledgement) and I’ll give concrete implementation steps for that approach. Sources: The ReAct Pattern for Voice Agents and How AI Agents Think, Act, and Respon... | Function tool definition | LiveKit Documentation | Bringing AI avatars to voice agents | LiveKit | Agent Builder | LiveKit Documentation, with interactive elements