# AI in your product

> Read the AI in your product pages: how your AI features are doing, what they cost, which runs and tools fail, and your MCP server's tool calls.

Section: Each part of Relia · Page: https://tryrelia.com/docs/ai-in-your-product · Last checked against the product: 2026-11-01
Video walkthrough: https://youtu.be/szMNhbJQauU

**AI in your product** watches the AI features inside your product: the calls your code makes to model providers, the tools your AI uses while it works, and the tools your own MCP server offers to AI assistants. It is not about Claude Code or Cursor connecting to Relia; those are on your Profile ([Tokens and connected tools](https://tryrelia.com/docs/tokens-and-connected-tools.md)).

## What gets recorded

With the usual setup (`npx relia-sdk@latest init`, or `python -m relia`), and nothing else to add:

- **Model calls**: each call to OpenAI, Anthropic, Google Gemini, OpenRouter, Groq, Mistral, DeepSeek, Together, xAI, Perplexity, Fireworks or Cohere becomes a run with the model, tokens, estimated cost, how long it took and whether it failed. Streamed answers included. In Node.js that covers the `openai` and `@anthropic-ai/sdk` packages, the Vercel AI SDK and plain `fetch`; in Python the `openai` and `anthropic` packages and anything using `httpx` or `requests`; in edge functions, calls made inside `withRelia`.
- **Tool calls**: when your AI calls its own tools (like `search_orders`), each one is a step of the run, with its time and whether it failed.
- **Your MCP server**: if your product offers tools to AI assistants with the official MCP SDK (Node.js or Python), every tool call is recorded: the tool, which assistant called it, how long it took and whether it failed. What the assistant asked and what the tool answered are never recorded.

Model calls and MCP tool calls need `relia-sdk` 0.8 or later on npm, or 0.6 or later on PyPI.

> **Tip:** watch [Monitor OpenAI, Anthropic and Gemini calls in your app with no code](https://youtu.be/xcQ2aAHnrRM) (2 min).

## The four pages

Open **AI in your product** in the sidebar, under **Your product**.

### Agents

At the top: how many AI features you run, how often they ran, and what they cost. **Fleet Success Rate** is the share of runs that worked; the target is 96 % or more. Below, each AI feature is one row with its runs, speed, success rate and tokens, marked **Healthy**, **Degraded** or **Failing**. Open one for its trend and which providers and models it uses.

### Runs

Every time the AI answered: the question, how long it took, the tokens and what it cost. Choose **Errors Only** to see what failed, for example a provider answering "rate limited". Open a run to see each model call and tool call in order.

### Tool calls

The tools your AI uses: how often, how fast, how often they fail. Below it, **Your MCP server** lists each tool your server offers with calls, failures and speed, and which AI assistants called it. Click a tool to see its latest calls. If you have no MCP server, this part is empty.

### Agent issues

Relia flags AI that loops, retries a failing step, runs slow, or uses far more tokens than usual.

## When Relia raises an incident

Relia reads every run of the project, including runs from background jobs and scripts. It raises an incident when runs fail or get slow, when spending jumps (at least $1 and 3 times the usual for the same length of time), or when an MCP tool starts failing or slowing down. A run inside a failing request joins that request's incident, so you see one story.

> **Tip:** watch [AI feature failing or costing too much? Monitor LLM calls and MCP tools](https://youtu.be/OCDIWdXLIQs) (2 min).

## Good to know

- Costs are estimates from public list prices. A model Relia doesn't know shows no cost rather than a guess.
- Prompts and answers of model calls are kept with e-mail addresses, phone numbers and card numbers masked.
- Each AI run counts 5 units of your plan's monthly pool. MCP tool calls cost no extra units.
- Turn off model-call capture with `RELIA_AI_CAPTURE=0`, and MCP recording with `RELIA_DISABLE_INSTRUMENTATIONS=mcp`.

## Next

- [Monitor AI calls with no code](https://tryrelia.com/docs/ai-calls-setup.md)
- [AI feature failing or costing too much?](https://tryrelia.com/docs/ai-feature-failing.md)
- [Privacy](https://tryrelia.com/docs/privacy.md)

---
All Relia docs for AI assistants: https://tryrelia.com/llms.txt (index) · https://tryrelia.com/llms-full.txt (everything in one file)
