# Ask odo

> Chat with Claude inside odo about your numbers: where to find it, what it reads, the changes you approve, who has it, its limits and privacy.

**Ask odo** is a chat with Claude inside odo. Ask about your numbers in plain words: Claude looks them up in odo, answers with the numbers behind it, and says where in odo to see them for yourself.

## Where to find it

- **Ask odo** in the bar at the top of the Overview, Revenue and Traffic dashboards.
- **Ask odo** in the dock on the live counter (under **More** on a narrow screen).
- On something odo already shows you: the sparkle button beside each of the **Next moves**, **Ask about this** on each next step in **odo’s review**, and **Ask odo about this post** in a post’s sheet. Your question then comes with what odo shows about it, so you needn’t explain it again.

A new conversation opens with a few questions to start from, picked for what the project has. Good ones to ask:

- *Why did sales dip on Tuesday?*
- *Which post made the most money this month?*
- *Where am I losing buyers?*
- *What should I test next?*
- *What did my last site change do?*

While Claude works, each lookup shows as a short line (*Looking at traffic*, *Comparing segments*), then the answer follows. **Stop** ends an answer early, **New conversation** starts afresh, and **Conversations** lists your earlier ones in this project.

## What Claude reads

Claude reads only what odo has for the project on screen, and only what you can see there. In **Global**, that’s all of your own projects together, and a project you manage for someone else when you name it.

It looks things up with the same tools as an [AI assistant connected over MCP](https://odo.gg/docs/mcp): earnings and sources, visits and their journeys, pages, funnels, heatmaps, experiments, site updates and odo’s own **Next moves**. The [tool reference](https://odo.gg/docs/mcp-tools) lists every one. Claude is told to answer from what those tools return, not from general knowledge or a guess, and:

- To quote rates with how many people they rest on, and not to call a difference real when it could be chance.
- To say which money odo read from the platform itself (verified) and which was self-reported.
- To say where in odo to look, by the names odo’s screens use: *Traffic → What they do → Funnels*.
- When something isn’t set up (no website measured, sales not traced to visits, no goals), to say what that keeps it from answering, and what to set up.

## Changes you approve

Besides reading, Claude can do three things in odo: note a [site update](https://odo.gg/docs/site-updates), make an [experiment](https://odo.gg/docs/experiments), and change one (start, pause, resume, stop, ship or archive it). It never does one on its own say. It proposes it, and odo shows the change in full first: what it is, a line on what it does, and every value it will use.

- **Approve** makes exactly that change, and Claude carries on from the result.
- **Cancel** tells Claude no. Asking something else instead says no too.

Claude proposes one change at a time, and only when you asked for it or agreed to it. Anything else (your site’s code, sources, settings, goals and funnels) stays yours to change: Claude says what to change, and where.

## Who can use it

Ask odo comes with the [Growth plan](https://odo.gg/docs/plans), and with the free trial, which has everything Growth has. On Starter, it says it comes with Growth.

The owner, admins and managers of a project can use it. Viewers can’t, and shared links and public pages never show it. More on roles in [your team](https://odo.gg/docs/team).

## Limits

| Limit | How much |
| --- | --- |
| A question | Up to 4,000 characters |
| Lookups for one question | 12, then Claude answers with what it found |
| Questions in one conversation | 40, then you start a new one |
| Questions a day, for the account | 100 |
| What Claude reads and writes a day, for the account | 1,000,000 tokens |

The account is the project owner’s: everyone using Ask odo in their projects shares its questions and tokens. A day is the last 24 hours, not the calendar day. With 10 questions or fewer left, the box says how many; once a limit is reached, it says when Ask odo opens again.

## Privacy

- **Your conversations are yours.** Each one belongs to you, in the project it’s in. Nobody else sees it, the project’s owner included. **Delete** beside one in **Conversations** removes it and everything in it.
- **Each conversation starts fresh.** Claude doesn’t carry anything over from your other conversations.
- **Text from your site is data, not instructions.** Page paths and titles, post captions, event, form and experiment names, error messages, site update notes and anything else from your pages or your visitors’ browsers reach Claude as data. odo tells it never to follow that text, nor let it change what it looks up, what it proposes, or what it tells you to add to your site.
- **Claude can be wrong.** Before you act on a number that matters, check it on odo’s own screens: Claude says where to look.

## More

For the reads odo writes by itself (a verdict on each post, the weekly review, experiments’ read-outs), see [odo’s analysis with Claude](https://odo.gg/docs/analysis). To ask from Claude, ChatGPT or a coding assistant outside odo, [connect it over MCP](https://odo.gg/docs/mcp). The [prompts to try](https://odo.gg/docs/prompts) are written for those, and most work in Ask odo as they are.