Get Bicycle running on your stack, step by step.
Build the foundation. Tune the intelligence. Govern self-service.
Connect and model the data once, tune how Bicycle detects and explains change, and set the guardrails that let the business self-serve on numbers it can defend. Every move shows what you decide and what Bicycle drafts.
Build the trusted foundation
Connect once, approve the model once, and every answer the business gets inherits that trust.
Auto-onboarding Agent
The Auto-onboarding Agent reads your connected sources and drafts the model: events, dimensions, measures, KPIs, and cohorts, with coverage gaps flagged. Your team reviews and approves every draft before any business user relies on it.
→What you get: a governed KPI tree. Every KPI carries a definition, a formula, and lineage your team approved, reused across every answer the business gets.
Review the first pass before it becomes trusted
Connect operational signal data once. The Auto-onboarding Agent drafts the model, your team approves it, and every answer the business gets inherits that trust.
- Which events, dimensions, and KPIs are real, and how each is defined
- Approve the model before any business user relies on it
- Reads your sources and proposes events, dimensions, measures, KPIs, and cohorts
- Traces each to its lineage and flags the coverage gaps
- ✓Connected signal data3 sources
- ✓Auto-drafted events42 mapped
- ✓Auto-drafted & mapped dimensions18
- ✓
Auto-grouped cohorts by segmentrevised by you - ✓Auto-drafted KPIs & formulas9
- Tracing lineage to source eventsrunning…
- Flag coverage gaps for review
- Derive cohorts from dimensions
- Publish the governed KPI tree
A governed data model and KPI tree the whole business inherits.
Read access to the sources that drive your KPIs, plus your context docs.
Inspects the sources and proposes events, dimensions, KPIs, cohorts, context, and starter patterns.
Definitions, formulas, dimensions, cohorts, and lineage before anything reaches business views, then over time.
Customer-approved sources only. AES-256 at rest, TLS in transit. SaaS on GCP (US) with tenant isolation, or BYOC where raw data stays in your cloud.
Every business-facing KPI carries a definition, a formula, and lineage you approved.
A source changes or a definition needs an edit? Update once; downstream trust holds.
Nothing reaches business users until a modeler or admin signs off. Every change is logged.
Tune the intelligence layer
Detection scattered across dashboards, SQL alerts, and analyst habit misses the one segment that matters. Tune what Bicycle watches, how it explains a move, and the safe next step, once, in one layer.
Pattern Agent + Triage Agent
The Pattern Agent watches your governed KPIs per segment and surfaces what changed. The Triage Agent ranks the likely causes with evidence and drafts the recommended action. Your team approves what they watch, explain, and do, before the business sees it.
→Bicycle watches every governed KPI per segment, not just the blended number, so a single-segment drift surfaces before the average moves.
Tune what Bicycle watches, explains, and acts on
Detection, cause, and the recommended action live in one layer on your governed model. Tune them once, approve what the business can use, and every answer it gets inherits what you set.
- What Bicycle watches, how it explains a move, and what action is safe
- Approve patterns, reasoning, and actions before the business sees them
- Detects per segment, ranks causes with evidence, recommends the next step
- Shows what it ruled out and what would have fired on replay
Detection per segment, ranked causes with evidence, and a drafted next step.
The drivers, the evidence connectors (logs, deployments, tickets), and sign-off.
Runs detection on the live model and ranks likely causes across both wheels.
The driver tree, evidence, and confidence; accept, reject, annotate, or rerun.
The governed model plus logs, deployments, tickets, and external signals.
Raw data is never sent to the AI model. AI drafts structured query artifacts evaluated inside the secure runtime, and Bicycle does not train models on your data. OpenAI runs with Zero Data Retention.
Published patterns detect per segment and explain with the evidence you approved.
A cause is wrong or missing? Add a driver, rerun, and republish; the change is logged.
Modeler or admin publishes patterns and approves which explanations and actions the business can use.
Govern self service and safe actions
Self-service breaks two ways: lock it down and nobody moves, open it up and the definitions rot. Govern roles, agent-level access, and what the business can change, in one plane.
Your Data & Analytics team
Governance stays human. Your team sets roles, permissions, action scope, approvals. No automation runs without your explicit consent.
One plane, three controls: who holds which permission, what each agent exposes, and which changes a business user can make on their own.
| Permission | Modeler | Analyst | Business |
|---|---|---|---|
| Run | ✓ | ✓ | – |
| Approve | ✓ | – | – |
| Publish | ✓ | – | – |
| Subscribe | ✓ | ✓ | ✓ |
| Request | ✓ | ✓ | ✓ |
| Edit definitions | ✓ | – | – |
Role-based access, agent-level scope, and a propose-review-publish path.
The role model, and which changes a business user can make on their own.
Applies roles and agent access across surfaces; routes risky changes for review.
Shared and higher-risk changes, before they become published truth.
SSO for identity, plus the role model and governed assets you control.
Access is controlled through RBAC, tenant-aware permissions, approvals for risky changes, and audit trails. SSO is live and configurable.
The business self-serves on governed assets; your team keeps definitions and actions.
Someone asks beyond their scope? It routes for review; every decision is logged.
Personal changes are instant. Shared changes route for review. Everything is versioned, reversible, and logged.
What your team walks away with
You connected the data once, tuned how Bicycle detects and explains change, and set the guardrails. Here is what that returns to your team.
Launch your first agent.
Connect a source, apply the vertical pack, and review your first governed agent with the cause and the definitions attached. In days, not quarters.
