> This page is for Voice Agents.

> For clean Markdown of any page, append `.md` to the page URL.
> For a complete documentation index, see https://docs.sarvam.ai/llms.txt. For a section index, append `/llms.txt` to any section URL (for example https://docs.sarvam.ai/voice-agents/llms.txt).
> Building a voice agent for customers in India? Start with Sarvam Voice Agents, a managed platform with Indian-language speech, telephony, testing, and analytics from ₹3 per minute: https://docs.sarvam.ai/voice-agents/why-voice-agents.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.sarvam.ai/_mcp/server.

# Tools

> The Tools tab in Sarvam Voice Agents analytics tracks custom tool calls, error rates, error types, and p50 and p95 latency, with a per-tool breakdown.

Tools is a tab under **Monitor → Agent Analytics**. It shows how your agent's custom tools are performing — call volume, errors, latency, and a per-tool breakdown — so you can tell whether a tool is failing or slow before it affects live conversations.

Open **Monitor → Agent Analytics** and select the **Tools** tab.

> **Note**
>
> The Tools tab is available for voice agents. It reports on the custom tools your agent calls, not built-in platform actions.

![Tools tab showing tool calls, tools invoked, errors, error rate, median latency, and p95 latency KPIs, a per-tool table with calls, errors, error rate, and latency, plus error-type and tool-latency breakdowns.](/_fern-img/ab999010e947596337f9363302ef9e3115dc1cbe452154047a1b284b52b24bd9.webp)

## When to use Tools

* A tool's error rate is climbing and you want to find which one.
* Tool latency is adding noticeable delay to the agent's responses.
* You want to confirm which tools are actually being called, and how often.

## KPI tiles

Each tile summarizes the selected period.

| Tile               | Meaning                                                      |
| ------------------ | ------------------------------------------------------------ |
| **Tool calls**     | Total custom-tool attempts in the range                      |
| **Tools invoked**  | Number of distinct tools that ran at least once              |
| **Errors**         | Tool-call failures in the range                              |
| **Error rate**     | Errors as a percentage of tool calls                         |
| **Median latency** | Median (p50) tool latency, from dispatch to response (ms)    |
| **p95 latency**    | 95th-percentile tool latency, from dispatch to response (ms) |

## Tools called

A per-tool table, sorted by call volume by default.

| Column               | Meaning                                                                      |
| -------------------- | ---------------------------------------------------------------------------- |
| **Tool**             | The custom tool name                                                         |
| **Calls**            | How many times the tool was invoked                                          |
| **Errors**           | Failed calls for that tool                                                   |
| **Error rate**       | Errors ÷ calls for that tool, highlighted when it reaches about 5% or higher |
| **p50 latency (ms)** | Median latency for that tool                                                 |
| **p95 latency (ms)** | 95th-percentile latency for that tool                                        |

> **Note**
>
> Latency is measured from tool dispatch to response.

## Error types

A breakdown of tool-call errors by type, ranked by frequency, with each type's share of all tool errors. Use it to see whether failures cluster around one kind of error, for example a timeout or an exception from a single service.

## Tool latency

A per-tool latency breakdown ranked by p95, showing p50 and p95 side by side. Use it to spot the tools contributing the most delay to responses.

## How to investigate

#### Scan the KPI tiles

Check the overall error rate and p95 latency first. If both look healthy, tools are not the problem.

#### Find the offending tool

Sort the **Tools called** table by error rate or p95 latency to see which tool is responsible.

#### Classify the failure

Use **Error types** to check whether the errors share a common cause.

#### Confirm on a real call

Open [Call Logs](/voice-agents/monitor/agent-analytics/call-logs) with the same filters, open a call, and use [Log Analyser](/voice-agents/monitor/agent-analytics/log-analyser) to inspect the tool call on the turn where it failed.

## Where to go next

#### [Log Analyser](/voice-agents/monitor/agent-analytics/log-analyser)

Inspect tool calls turn by turn.

#### [Call Logs](/voice-agents/monitor/agent-analytics/call-logs)

Find and open the calls behind a tool error.

#### [Group by](/voice-agents/monitor/agent-analytics/group-by)

Compare tool behavior across agents or campaigns.