so, when I send conversations_update messages to my Elevenlabs Agents, it sometimes starts to call

[blazing summit]

AI-generated Answer for _podden

Short answer: it’s expected. You’re sending the “contextual_update” client event (the documented event name), and the platform records these updates in the conversation history as tool-call entries. They are not real client tools and do not require a client_tool_result; they’re just how ElevenLabs surfaces context changes.

What the docs say:

• The client-to-server event is type: "contextual_update". It’s “non‑interrupting” and “updates are incorporated as tool calls in conversation history” and “do not require a direct response.” (elevenlabs.io)
• In the WebSocket API reference, “contextual_update” is a first-class message type distinct from real “client_tool_call” events (which are the only ones you should answer with client_tool_result). (elevenlabs.io)

So if you’re seeing a tool name like “contextual_update” (or a misspelling of it) in logs, that’s just the history representation of your contextual update—not a missing tool on your side.

Sources:
WebSocket | ElevenLabs Documentation
Agent WebSockets | ElevenLabs Documentation

[pale summit]

so in recorded and analysed conversations I get this tool called as failed and the LLM is responding like "Oh there seems to be a problem" which is not ideal to say the least. I could just create an empty tool but I want to know if I have found a bug or if there's a problem on my side