Analytics

Shipped

Your Turbobase instance ships with a built-in operational dashboard on the app’s home page, so you can see at a glance whether your sync is producing data and whether the instance is healthy — without wiring up an external tool.

The home dashboard

When you open your Turbobase app, the home page leads with the proposal composer — the same box, and Inputs drawer, as the Development pipeline — so the first thing you can do is start a proposal. It stays disabled until the initial sync finishes.

Below it, administrators get the dashboard proper, and it comes in three views you switch between at the top of the page. One page, three questions:

• Executive (the default) — what the instance is producing. Proposals by status, objects in the catalog, open Inputs, what has been published to your team, customization risk by family, open Inputs by detector, and the improvement pipeline from proposal through complete and realized to live.
• Operations — what is costing somebody time. What needs a person, manual edits today, what is waiting on approval, manual re-keying against automated writes, which failure classes cleared themselves over the last 30 days, and where the re-keying lands by record type.
• Developer — how the instance is behaving. Proposals in flight, solutions open, read p95, active extensions, read latency over the last hour, AI tokens by operation, and each extension’s fires, rejections and errors.

Each view loads only its own panels, and a panel that can’t be read shows its own empty state rather than taking the page down with it. Recent proposals and releases are listed alongside, and each proposal links into Studio, the proposal-authoring canvas.

The composer leads for everyone; the three views are an administrator view. Non-administrator users still get the dashboard chrome — navigation and the chat sidebar for Pilot, your AI assistant — without those operator panels.

Agents, services, and AI usage

Health of the AI agents and background services lives in the account menu — open it and expand Agents & services:

• AI availability — a headline showing whether AI is available right now, with the reason when it isn’t (budget exhausted, quota, rate-limited, or provider overload).
• Agent status — a live status row for each agent.
• Services — a collapsed drill-down of process health. You don’t need to open it to know something is wrong: the account avatar shows a red dot whenever anything is down.
• Token usage — application-wide input and output token totals over 24 hours, 30 days, and lifetime, broken down by tier: Frontier, Chat, Fast, and Other for a model that fits none of the three.

From the usage block, the breakdown button opens the full AI usage modal: per-operation tokens, cache reads and writes, and a stacked usage-over-time chart, all grouped by tier. Zooming in refetches the visible window at finer time buckets — genuinely more detail, not stretched pixels. Usage is reported as token counts by tier; the underlying AI provider and dollar cost aren’t surfaced.

How the metrics are produced

The instance samples its own internal metrics every minute and rolls them up into per-minute buckets, computing throughput, error rates, and latency quantiles (p50 / p95 / p99). High-cardinality dimensions like per-route breakdowns are capped so a misconfiguration can’t explode the dashboard — traffic past the cap collapses into a single __other__ bucket so the totals stay accurate.

Querying your own data

For analytics over your business data itself, query the canonical data model directly over Postgres or PostgREST, or ask Pilot. Everything you’ve mirrored from your sources is available as typed, queryable tables.