Skip to main content

Usage dashboard ENTERPRISE

getdeepstate.app/usage shows how much DeepState AI your team is using. Sign in with the master key.

What it shows​

Choose Last 7 / 30 / 90 days. Then:

  • Totals - total tokens, input tokens, output tokens and number of requests across all seats.
  • Daily token usage - a bar chart by day, so you can see the shape of the work: the spike when a new investigation started, the quiet weekends.
  • Usage by seat - one row per seat with the holder's email, input and output tokens, total, and request count. Sort by clicking a column header.

Only usage through DeepState AI is counted. Seats that have switched to their own provider key (Claude, OpenAI, OpenRouter, Ollama…) show as zero here even when they are busy - their usage is on their own account.

Figures update within a few minutes of the requests being made.

Tokens and budget​

A token is a fragment of text, roughly three-quarters of a word. Every AI request sends tokens in (your prompt, the board context, tool results) and gets tokens out (the reply). Output tokens cost several times more than input tokens with most models.

Each seat has $8 per month of DeepState AI, resetting monthly. The dashboard shows tokens rather than dollars because the price per token depends on the model each seat chose; as a rough guide, a Sonnet-class model turns $8 into a few million input tokens or a few hundred thousand output tokens. When a seat hits its limit, its DeepState AI requests are declined with a clear message until the reset; the holder can carry on immediately by switching to a personal provider key in Settings → AI Provider.

What drives usage​

  • Autoresearch sessions: each round is a full turn; fifteen rounds on a broad topic is the single most expensive thing in the app.
  • Reasoning depth set to high.
  • Boards with many active feeds and the AI relevance filter on - many small summarizer calls.
  • Large context: selecting dozens of nodes before asking, or very long conversations that have never been compacted.

Cheap habits: a Haiku-class summarizer model, low reasoning for routine turns, compacting long conversations, and Plan mode for questions that don't need the board changed.

Privacy​

The dashboard shows counts and timestamps only. Neither the administrator nor DeepState can see prompts, responses, node content or which boards a seat is working on. See Security & privacy.