Insights and cost · Faundr Docs

Project dashboard

Insights and cost

Insights is the first screen of every project: a dashboard with the session context, usage limits, cost, tokens and project health. It shows how many tokens Faundr uses and saves, what your sessions cost in dollars and how the project's health has changed since you turned Faundr on. There's no command to run. The numbers arrive on their own.

Dashboard

The tab that opens with the project.

Session context, 5-hour window and Weekly limit: how much the last session used, and when each window resets. The card is highlighted at 80%. In Claude Code, these numbers come from the Faundr status line in the terminal. Without the status line, the card explains how to turn it on (!faundr statusline-install). Codex has no status line, so there they come straight from the conversation. The limits belong to the account of whoever ran the session. They aren't tied to the project.
Cost this month, the 5-hour window in dollars, Faundr's share (what it used and what the code map saved) and Issues fixed. Click any of them to open the tab with the details.
Cost per day and Tokens per day for the last 14 days. Hover over a bar to see that day.
Health now: open issues in each area, and whether things improved since the first snapshot.
Recent sessions, with the cost of each one.

With two people on the project

Context, the 5-hour window and the week are always yours. These limits belong to your AI account. Right below, a line shows each teammate's limits (from their last session, up to a day old).
Team or Just me. The switch at the top flips cost, tokens and recent sessions between the whole project and just your sessions. Project health always covers the whole project.
The 5-hour window in dollars is per person: on the Dashboard, in the Cost tab and in the budget warning the agent receives. Your co-founder's session never counts toward your window.
Who spent what: session lists show each person's name, and the Cost tab has By person, this month.

How the numbers get here

At the end of every response, the plugin adds up the tokens in your agent's conversation and sends only the numbers: tokens per model, tools used, how much Faundr added to the context and what the end-of-response check cost. The measurement reads the conversation on your computer and sends only the counts. The conversation text stays behind.
The terminal status line stores the context and usage limits your agent shows there, and the end of each response sends them along (just the percentages and reset times). In Codex, the same numbers are read from the conversation, on your computer.
On every check (security, quality, design, tests or errors), Faundr saves a snapshot of the project's numbers whenever one of them changed. The same happens when a connected skill records what it found.

Tokens

The headline answers the main question: "Faundr used ~N% of tokens".

The Where the tokens went bar splits the total into three parts:

Part	What it is
Agent work	What the agent spent doing what you asked.
Faundr context	The session-start summary, per-file rules and notices. This is an estimate.
End-of-response check	What the end-of-response requests cost (proof, handoff note, design review…).

The counters show total tokens, Faundr context, the end-of-response check, how many times the check fired and how many code map queries ran.

The What the code map saved card estimates how many reading tokens the map queries saved: what the agent would have spent opening the files the map pointed to. Treat it as a ceiling. It isn't an exact measurement.

Cost

What your sessions cost in dollars, calculated at the API price of the model used (Anthropic's price list).

If you use AI through a subscription (Pro or Max plans), you don't pay per token. In that case, cost is useful for comparing sessions and features, and for seeing how much of your plan you're getting.

This month: spend for the month, and the budget if you set one.
Your 5-hour window: your spend in the current usage window. Pro and Max plans cap usage in 5-hour windows, one per account.
Sessions measured, Average per session and Total measured.
By feature: what it cost to build each feature.
5-hour windows: spend in each window, including the one that's open right now.
Most expensive sessions.

Cost is based on the price list for Anthropic's models. A model with no known price (Codex models, for example) is left out of the total, and the Cost tab lists it under No known price (not counted). Its tokens still count in the Dashboard, Tokens and By session tabs.

Budget

Set a limit per month (shared by the team) and/or per 5-hour window (applies to each person), in dollars, and the percentage at which to warn you (80% by default). Once spend passes that point, the agent gets a warning at the start of the next session and the dashboard shows an alert. Leave both values empty to remove the budget.

Project health

How the project has changed since the first snapshot:

how many issues have been fixed since then, and how many disappeared in the same session they appeared in;
one card per area (security, quality, design, errors, tests), with the current number and a timeline;
whether each area improved or got worse since the start.

Problems found by connected skills (the ones tagged via /name) count right alongside Faundr's own.

The timeline shows the project since Faundr was turned on, but it doesn't prove the improvement came from Faundr. The team also fixes things on its own.

By session

One row per conversation: tokens, responses, prompts, Faundr's share, how many times the end-of-response check fired, how many map queries ran and whether the context was compacted partway through. When you're viewing the whole team, each row shows whose session it was. Search by person, branch or date.

Using less of your plan

Economy mode: under Settings → General, it makes the quick design audit and the Overview and Stack updates run only when you ask. The card shows what end-of-response checks cost over the last 30 days. See Settings.
Before generating anything with AI: /faundr:setup separates what's free (the code map, the Stack read from your files, the checks) from what uses your plan (Overview, reviews, design.md…), and estimates each item in tokens, as a share of your 5-hour and weekly limits, and in dollars at API rates. The estimate starts from the size of your code and switches to real measurements once a skill has actually run.

Checking what gets sent

Run !faundr usage --debug in your agent to see, as JSON, exactly the numbers Insights would receive from your most recent conversation.

Commands

Insights is the project's home screen: session context, 5-hour and weekly windows, cost and tokens per day. It shows how many tokens Faundr spends and saves, how much your sessions cost (per month, 5-hour window, feature and session, with a budget and alerts) and how the project's health has changed since Faundr was turned on. No command needed: the numbers arrive on their own. Run these commands in your AI agent, from the folder linked to the project.

At the end of every response, the plugin adds up the session's tokens (numbers only, never the text). On every security, quality, design, tests or errors check, Faundr saves a snapshot of the project's numbers whenever one of them changed.

!faundr graph-query "<question>"
Asks the code map and gets back only the part that matters, instead of the agent opening several files. At the end of the answer you see what it cost and what the cited files would have cost: that is the savings Insights adds up.
e.g. !faundr graph-query "how does login work"
!faundr usage --debug
Shows the usage numbers Insights would receive from the latest session, to check what is sent.

PreviousLaunch and privacy NextWhat Faundr does on its own