Skip to content
Agent Orchestrator Docs
Esc
↑↓navigate↵open⌘Jpreview
On this page

Usage & Cost

Track real token usage and cost across your Workspace — not estimates. Administrator-only page.

Track real token usage and cost across your Workspace — not estimates. Administrator-only page.

Enterprise edition. Usage & Cost is part of the Enterprise edition; the Community edition does not show it.

Usage & Cost

Overview

The Usage & Cost page shows:

KPI Tiles (Top Row)

Metric Description
Total Cost Sum of all run costs in date range
Total Tokens Input + output tokens
Total Runs Count of completed runs

Default range: Last 30 days (configurable via date pickers).

Breakdown Tabs

Switch between four dimensions:

Usage Breakdown by Model

Tab Groups By Use Case
By Runtime Runtime name Which execution environment costs most
By User User email Who triggered the most expensive runs
By Agent Agent name Which agents consume most resources
By Model Model name Compare model costs (e.g., Claude vs GPT)

Each tab shows a table:

  • Label — Runtime/User/Agent/Model name
  • Cost — Total USD
  • Tokens In — Input tokens
  • Tokens Out — Output tokens
  • Runs — Count

Sorted by cost descending.

Date Range Filter

Toolbar at top:

  • From — Start date (defaults to 30 days ago)
  • To — End date (defaults to today)
  • Clear Filters — Reset to default 30-day range

Data refreshes automatically on filter change.

What’s Tracked

Per run, the daemon captures from opencode:

  • tokens_in — Input tokens (prompt + context)
  • tokens_out — Output tokens (completion)
  • cache_read_tokens — Cached input tokens
  • cache_write_tokens — Cached output tokens
  • reasoning_tokens — Reasoning/effort tokens (if model supports)
  • cost — Calculated USD cost
  • model_used — Model identifier
  • triggered_by — user / agent / scheduler / system

Stored in runs table (V29 migration).

Data Sources

Source What It Captures
Daemon opencode step_finish events → token/cost per run
Backend Aggregates on /complete and /fail callbacks
PostHog Server-side product analytics (specs/020) — business events

Permissions

  • Administrator — Full access to Workspace’s usage
  • System Administrator — All Workspaces (via instance settings)
  • Members — No access

Product Analytics (PostHog)

Separate from Usage & Cost — tracks business lifecycle events:

Event Category Examples
Auth user_signed_up, user_logged_in, sso_login
Workspace workspace_created, member_added, member_role_changed
Agents agent_created, agent_updated, skill_assigned
Skills skill_created, skill_published, skill_imported
Tasks task_created, task_assigned, task_status_changed
Runs run_started, run_completed, run_failed
Deliveries delivery_proposed, delivery_approved, delivery_rejected
Squads squad_created
MCP mcp_server_created
Runtimes runtime_registered

Privacy: distinct_id = user UUID (matches frontend posthog.identify), daemon events use triggered_by. No PII in event properties.

Configuration

Instance settings (System Admin only) → Observability tab:

  • Analytics Enabled — Toggle PostHog event capture (default: on)
  • OTel Logs Enabled — Toggle log shipping (default: on)
  • Both require env vars (POSTHOG_API_KEY, OTEL_EXPORTER_OTLP_LOGS_ENDPOINT)

Recent Updates

  • Usage & Cost Tracking — Real token/cost data per run, workspace totals, breakdown by Runtime/User/Agent/Model
  • Product Analytics — Server-side event capture to PostHog for business lifecycle events
  • Date Range Filter — Configurable range with 30-day default