See every tool call your LLM agent makes
October 27, 2026 · release
@voicethere/agent 0.17.0
When an LLM agent answers wrongly, the first question is usually what the tool returned. Until now you could only guess from the reply. With @voicethere/agent 0.17.0 every tool call writes one line to your project's agent logs: the tool, where it came from, how long it took, whether it failed, and what went in and came out.
The LLM tab lists these calls newest first and lets you filter by tool or session. Open a row to read the input and output. On the session page, the transcript shows each call between the caller's turn and the agent's reply, so you can read a turn from question to tool to answer.
The agent redacts before it logs. Secret values from your AGENT_ secrets, values under keys like password or token, and personal data such as email addresses, phone numbers, card numbers and IBANs are replaced with markers like [redacted:email]. Each of input and output is cut at 2048 characters, and the row tells you when that happened. If you need the raw values while you debug, set toolCallLogs to "full" in voicethere.llm.json; secret values stay removed even then. Set it to "off" to log no calls at all.
Try it
- Docs: Tool call logs
- Dashboard: open a project, go to the LLM tab and look under Tool calls
- Package: update
@voicethere/agentto 0.17.0 if you copied the loop into your code