Ingest
Telemetry-shaped inputs
Batch history plus steady-state appends so AI questions stay current without manual re-exports.
Ask what changed in your product data—
from the AI tools you already use.
Bring warehouse tables and append-only events into a project-scoped query path. Row limits, time windows, and access checks apply on the server before results reach Cursor, Claude, or your own assistants—not after the fact.
Scroll for the walkthrough, or move the sliders in the demo—the JSON preview updates to match the guardrails you set.
Interactive policy preview
Sample events read-only demo
| ts | user | event | status |
|---|---|---|---|
14:02:01 |
alex@acme.co redacted | feature_used |
ok |
14:02:04 |
alex@acme.co redacted | checkout_start |
warn |
14:02:09 |
sam@acme.co redacted | api_error |
error |
Guardrails applied before query
When on, email-like columns are omitted from the preview payload below.
What the client receives after policy
Every new copilot or analytics assistant turns into a custom pipeline: one-off exports, fragile joins, and questions that are hard to keep inside the same access rules as your APIs.
Hiddenplane is for teams shipping product telemetry and warehouse slices into AI workflows—where scope, caps, and project ownership are enforced on the server, not left to whatever the model was handed in chat.
Flagship path
From live product data to governed answers
The flagship path we build toward: connect the signals you already collect, let ingestion and profiling stabilize the schema, expose bounded views to AI clients, and keep queries inside project scope with explicit limits.
-
1
Connect a source
Application DB, warehouse slice, or append-only events—with ownership tied to a project.
-
2
Ingest & profile
Ingestion and indexing paths surface metadata and health signals so questions start from known fields, not guesses.
-
3
Expose to AI safely
MCP egress and analytics APIs expose list-and-query flows with bounded fields and caps—so assistants use the same paths as your services, not ad-hoc uploads.
-
4
Ask in your client
Use the same prompts across Cursor, Claude, or internal assistants—against the same governed contract.
-
5
Trace & iterate
Tighten prompts and views using request IDs, traces, and workbench review—so what you ship to production traffic matches what you validated.
Walkthrough
See the loop end-to-end
Scroll through a captured UI sequence—each beat matches a stage in the flagship story: connect data, shape questions in the workbench, and reuse what your team has already validated.
Beat 01
Connect governed sources
Databases, files, and streams land in a project boundary—with profiling so prompts start from known columns instead of guessed names.
Beat 02
Shape questions in the workbench
Use chat and structured query flows to iterate on charts and tables—review and tighten before anything is treated as production-ready.
Beat 03
Workbench discipline
One place to chart, diff, and reproduce—so answers stay stable when teams change.
Beat 04
Reuse what you have validated
Keep approved analyses and datasets inside the same project boundary so teammates and AI clients hit governed views—not one-off exports.
Snap-stops are intentional — scroll, pause, scroll.
Outcomes
What “fast” means here
Design targets for the telemetry-to-AI loop we describe on this page—not a performance guarantee for every deployment.
0
First governed questions (goal)
0
Policy envelope reused across clients
0
Interactive guardrails in the hero demo
0
Representative surfaces: MCP tools and in-app workbench
MVP focus
Depth where flagship risk lives
We bias engineering toward ingestion, analytics, MCP, and the workbench—the documented MVP path. Broader marketplace and billing flows stay behind that until the core loop is proven.
Surface
MCP-ready views
List datasets, run bounded queries, return inspectable aggregates—not opaque blobs.
Guard
Quotas & caps
Row limits, time windows, and rate signals on high-cost paths—quota checks default to fail-closed in our services when limits cannot be verified.
Workbench
Iterate before you trust
Charts, diffs, and saved runs in the analysis workbench so reviewers can sign off before wider rollout.
Scenarios
Where teams feel the pain first
Three loops from our flagship use case—incidents, feature adoption, and plan or tenant load—grounded in the same project-scoped tables your stack already writes.
Incident questions (bounded)
Ask which features saw error spikes around a release—inside a time window, row cap, and column allowlist you set for model-facing queries.
Feature adoption without a BI queue
Compare modules across tenants and plans with aggregates that stay inside project ACLs—ideal for PMs pairing with an AI client.
Plan & tenant load
See which customer tiers drive cost or noisy traffic before you tune pricing—still governed, still quotable.
Trust
What we optimize for before growth claims
The social strip uses illustrative marks only—no implied customers or endorsements. Below is how we expect teams to evaluate real risk.
- Identity from auth, not headers. Tools and browsers do not get to widen scope by passing hints.
- Project-bound data. Queries reconcile to the caller’s authorized project before rows are read.
- Cost-aware paths. High-export and streaming features are designed for quotas—not silent runaway reads.
- Honest roadmap. Public marketplace and full billing are not the MVP bet; flagship work stays on telemetry and warehouse data → governed AI access.
Mission
Help teams connect live product and telemetry data to AI so they can see what changed in minutes—without spinning up a new data project every time, and without treating scope and caps as an afterthought.
How it works
Three moves instead of a science project
Same spine as the journey above—stated as the weekly decisions platform and product leaders revisit.
01
Connect telemetry
Land the tables or streams your product already emits—profiled so AI questions start grounded.
02
Ask inside guardrails
Let copilots draft SQL and charts, but only publish views that passed caps, ACLs, and review in the workbench.
03
Reuse with receipts
Package the approved analysis for teammates and AI clients inside the same governed views—expand to metering and marketplace flows only when that is the problem you are solving.
FAQ
Questions, answered
Scope, security posture, and how this differs from pasting a spreadsheet into a chat window.
How is this different from uploading a spreadsheet to an LLM?
Hiddenplane connects AI tools to governed product data instead of one-off files. Access checks, row caps, and time windows
run on the server before results reach the model.
What does MCP let my team do?
MCP lets Cursor, Claude, and internal assistants list approved datasets and run bounded queries through a structured API.
Your team can ask product analytics questions without copying raw data into chat.
Can we try it before connecting production data?
Yes. Start with a sample source, validate the AI analytics workflow, then connect a warehouse slice or append-only event
stream when you are ready.
How do you keep AI answers inside approved access?
Queries are tied to authenticated users, projects, roles, and server-side policies. The model only receives shaped
results that pass those checks.
Do you replace Datadog or our warehouse?
No. Hiddenplane sits between your warehouse, telemetry, and AI clients. You keep your existing systems while adding a governed
query layer for AI-assisted analysis.
How do you control query cost and usage?
Set row caps, time windows, and rate limits on high-cost paths. The goal is fast answers without silent runaway reads or
surprise AI data access.
How is access scoped inside a project?
Each dataset and query is checked against the project the user can access. Client-provided IDs are treated as hints, not
proof of permission.
Can we automate queries with API keys?
Yes. API keys can run governed queries through the same gateway paths as the UI, with the same quotas and authorization
checks.
Does this work for multi-tenant SaaS analytics?
Yes. Keep tenant and plan fields in your warehouse or event stream, then expose approved aggregates to AI clients inside
your project and policy boundaries.
Is there a free tier?
Public beta accounts are free for evaluation. Create an account to try sample data, run bounded questions, and review the
current source and query limits in-app.
Can we self-host or run in our VPC?
The public beta is hosted. If you need dedicated, VPC, or self-hosted deployment options, contact us so we can discuss
your security and data residency requirements.
Can we use the beta with sensitive production data?
Use sample, non-production, or redacted data until your security team reviews our controls, DPA, and subprocessors. Ask us
for the current compliance posture before using sensitive production data.
Connect data your AI tools can query safely
Create an account, open a sample dataset or your first source, and run a bounded question from the workbench or an MCP-aware client.
Built where high-volume product data is normal
Illustrative technology marks for visual scale and familiarity—they are not customer logos, case studies, or endorsements.