Analytics that match the bill.
Token volume by model, P50/P95 latency, host load, error breakdown, plan usage. Scoped to assistant, team, or org. Trend windows that compare against the prior period.
Token spend by day
model: claude-sonnet-4-6 + codexNumbers that show up before the invoice does
Glueprint already knows what the agent did — analytics is the same data, rolled up. No third-party telemetry; no agent reads exfiltrated to a vendor.
Token spend, model mix, latency
Per-day token volume by model. P50/P95 latency. Estimated spend that survives a Stripe webhook outage. No surprise on the invoice.
Per-assistant and per-team rollups
Drill into a single assistant or pivot to team. The same KPIs, scoped to a smaller surface — easy to spot the routine that's burning tokens.
Host load and session counts
Which host is doing the work, which is idle. Session counts, error rates, and the noisy tools driving them.
Plan usage in one place
Hosts, seats, teams — used vs. limit. Soft enforcement: existing resources keep running on a downgrade; only creation is blocked when over limit.
Errors with context
Tool denials, run_command timeouts, failed approvals. Each row links straight back to the session that caused it.
30 / 60 / 90 day windows
Compare to the previous window. See trends across releases. Filter by host, assistant, team, or model with one click.
Bring Glueprint to your workflow.
Download once, run on your hardware, scale up to teams when you're ready. No credit card.