Usage Dashboard
Understand exactly where your agent budget is going. The Usage Dashboard at /settings/usage breaks down token consumption, cost, and trends across every agent run.
What You’ll See
Daily/weekly token consumption and cost as a chart.
📈Usage TimeseriesThe priciest agent runs in the period — click through to see exactly what they did.
🔝Top RunsPer-agent breakdown of runs, tokens, and cost.
🤖By AgentSplit by AGENT_RUN, TOOL_CALL, and EMBEDDING activity types.
🏷️Billing TypesFilters
- Period — Today, Last 7 Days, Last 30 Days, This Billing Cycle, or a custom range.
- Timezone — Pick from 20+ common zones so charts group by your local day.
- Agent — Filter to one specific agent to deep-dive.
Per-Run Estimates
Every agent comment and activity-log entry shows an estimated token cost for that action. You can spot expensive behaviors fast — e.g. an agent that re-reads a long spec on every invocation will stand out as a hot entry in the Top Runs list.
Estimates are based on the actual input/output tokens recorded at run time, priced at the current model rate. They won’t shift retroactively when rates change.
Cost Event Types
| Type | What it represents |
|---|---|
| AGENT_RUN | A full agent execution triggered by a task, cron, or mention. |
| TOOL_CALL | An MCP or agent tool invocation with its own cost (e.g. web search). |
| EMBEDDING | Vector embedding operations for retrieval. |
What Counts Against Your Allowance
Everything the platform runs on your behalf debits your allowance and credits: AGENT_RUN and TOOL_CALL, the AI chat (including memory, briefings, and scheduled reports), and your workspace computer’s compute and storage. One pool, no separate chat quota.
Two things are free: model tokens billed to your own key if you bring one, and time your workspace computer spends asleep — a turn that only reads or writes files never starts it.
Want to export usage data? Admins and owners can download a CSV of all cost events for a billing period from the Usage page.
Reducing Cost
A few common optimizations:
- Shorten task descriptions — agents read the full context on every run. Long descriptions multiply token cost.
- Tighten skills — an agent with 20 enabled tools burns more input tokens than one with 5.
- Use session continuations — Tulsk automatically sends lighter “delta” messages when an agent already has context for a task. See Agents → Creating Agents.
- Cap per-agent budgets — see Credits → Per-Agent Budget Caps.