Site administration
See growth and usage
You want to know if Source grows: how many people are active, how long they stay, and how much of the work agents do. The analytics page shows this from the usage events of the site, day by day.
6 min read
You need:
- To be a site admin.
- Analytics on your deploy. Set up analytics has the steps.
1. Open Analytics
Open Admin, then click Analytics in the side menu, under Activity. Or visit Analytics. Only site admins see the page.
2. Pick a range
Click 7 days, 30 days or 90 days next to the tabs. The default is 30 days. Each day is a UTC day, and the range ends today.
Tip
Each tab and each range has its own URL. You can send a view to another site admin.
3. Pick a tab
| Tab | What it answers |
|---|---|
| Overview | Who is active, and how long people stay. |
| Agents vs humans | How much of the work agents do, next to people. |
| Growth & retention | How sign-ups and other counts grow, and if people come back. |
| Traffic | Where people come from, the most visited repositories, and the web vitals. |
| Features | The top REST and MCP methods, the src versions, and the count of each server event. |
What each tab shows explains each card and chart.
Done
You now see who is active, how long they stay, how much agents do, and if people come back.
What each tab shows
Overview
These cards show who is active:
- People today, People in 7 days and People in 30 days. These three cards have their own windows, so the range does not change them.
- Visitors: the signed-out browsers in the range.
- Active repos: the repositories with at least one event in the range.
These cards show how long people stay:
- Sessions and Pages per session.
- Median session: the time from the first event to the last event of a session.
- Median time on a page: the time that a page was visible.
Below the cards, charts show the active people, active repos and sessions of each day. One more chart shows the new and returning people of each day. A person is new on the day of their first event, and returning on each later day with an event.
Top pages lists the most viewed pages, with the median time on each page. A page of a public repo shows its real path, for example /@TanStack/router/blob/main/README.md. Every other page shows its route pattern, for example /@{$owner}/$repo/blob/$.
Agents vs humans
- Browser sessions and Agent sessions: the sessions of people in the browser, next to the agent sessions that people started.
- People who used agents: the people who started at least one agent session.
- Factory runs: the factory runs that finished.
- MCP calls and src CLI commands.
Two tables split the API calls and the git operations (clone, fetch and push) by the kind of caller. A caller is a person (user), an agent (agent), a CI job (job) or a signed-out caller (anonymous). Charts show the same split for each day, and the browser sessions next to the agent sessions.
Growth & retention
Each count has two charts next to each other:
- A bar chart of the count of each day.
- A line chart of the running total.
The counts are:
- Sign-ups.
- Repos created, imported and forked.
- Merged pull requests.
- CI runs.
- Stars.
The running total starts at 0 on the first day of the range. So with 7 days, it counts only the last 7 days.
Git operations and API calls per day splits the work of each day into people and agents:
- People: the git operations of a person, plus the API calls (REST and MCP) of a signed-in user.
- Agents: the git operations and the API calls of an agent.
CI jobs (job), signed-out callers (anonymous) and API calls from before callers were recorded do not count. When the range has none of these calls, the card says so.
Retention by sign-up week has one line for each sign-up week of the last 8 weeks:
- A line shows the share of the people of that week who came back in each later week, from W+1 to W+8.
- Newer weeks have darker lines. The legend names each week.
- A week has no line when nobody signed up in it, or when its first later week has not started yet.
- A line stops at the last week that has started.
Weekly retention, under the lines, gives the exact numbers. W+1 is the share of the people of that week who came back the week after. W+2 is the share two weeks after, and the next columns continue the same way. A blank cell is a week that has not started yet.
Each week starts on Monday, in UTC. The range does not change the retention lines or the table.
Traffic
- Countries: the people and the visitors from each country.
- Referrers: the other sites that sent visitors.
- Top repos: the most visited repositories, with their visitors and views. Click a name to open the admin page of the repository.
- Web vitals: the 75th percentile (p75) of each metric. LCP is the time until the main content shows, in milliseconds. INP is the time until the page answers a click or a key, in milliseconds. CLS is how much the layout moves, and has no unit.
Features
- Top REST and MCP methods: the methods with the most calls, with their channel.
- src CLI versions: the commands that each version of
srcran. - Events: each server event, with its count and the number of people.
What the numbers mean
Who counts:
- A person is a signed-in account. The bot and the GitHub people of imported content never count.
- A visitor is a signed-out browser. Its id resets on each reload, so the visitor count runs high.
- A session ends after 30 minutes with no event. A signed-out visitor also gets a new session on each reload.
- Sessions on the Overview tab are browser sessions. Agent sessions are on the Agents vs humans tab.
Other notes:
- API calls from before callers were recorded show as
unknown. - A private repository shows its name to site admins, with a Private badge. When Source no longer has the repository, its id shows.
- Each top list shows at most 20 rows.
If analytics are not available
When ClickHouse does not answer, or its read secrets are missing, the page shows Analytics are not available and the reason.
- Make sure that
CLICKHOUSE_READ_USERandCLICKHOUSE_READ_PASSWORDare set. Set up analytics has the steps. - If the secrets are set, make sure that the tunnel is connected. Step 4 of Run the tunnel on the server has the command.
Note
The page needs a connection and is not saved offline.