Reference & help
Glossary
Every CHouse UI term — trust state, promise, blast radius, remediation credential, query shape — in one line each.
Terms used across CHouse UI and these docs.
| Term | Meaning |
|---|---|
| Agent | An AI client (Claude Code, Cursor, …) that uses CHouse UI over MCP or the API with a personal access token. Governed on the AI Governance page |
| Approval class | Whether a fix is low-impact (one approver, remediation:approve) or high-impact (two approvers, remediation:approve_high) |
| Baseline | What CHouse UI learned is normal for a table — write cadence and hourly volume — or for a query shape's latency |
| Blast radius | Everything downstream of an incident's root cause: tables, promises, jobs, saved queries and agents |
| Canary | A connection on a newer ClickHouse version used to replay your workload before upgrading |
| Collector | The server component that reads ClickHouse metadata on a schedule and stores the evidence. See How data is watched |
| Connection | A ClickHouse server CHouse UI talks to, with encrypted credentials. See Connections |
| Coverage | How many tables are protected by promises, by criticality. See Coverage |
| Criticality | critical, important or normal, learned from how much a table is read; can be pinned by hand |
| Data access policy | A named set of database/table rules, per connection, attached to roles. See Data access rules |
| Deployment (AI) | A named model configuration people use, pointing at a provider model. See AI models |
| Evidence | What the collector stored from system tables, which every status, root cause and suggestion is computed from |
| Fix / remediation action | One action from the closed remediation catalog, run only after approval |
| Health notice | A note attached to an agent's query result when a table it read has an open incident or is degraded |
| Incident | Something wrong with the data, a pipeline or the engine — open, acknowledged or recovered. See Incidents |
| Lease | A short-lived claim in the RBAC database that makes one pod do a piece of background work at a time |
| Lineage | How data flows between tables, from metadata and query_log. See Lineage |
| Maintenance window | The UTC window in which TTL and codec fixes may run (REMEDIATION_MAINTENANCE_WINDOW) |
| MCP | Model Context Protocol — how AI agents call CHouse UI's tools at /mcp |
| Notebook | A shared investigation record of AI findings, query snapshots, notes and fixes, exportable as a postmortem |
| Permission | A string such as logs:view that opens a feature. See the permission catalog |
| Personal access token (PAT) | A ch_pat_… token that lets the CLI, agents and scripts act as you. See Personal access tokens |
| Pipeline | Any way data enters ClickHouse — a view, queue engine, replication, writer or job. See Pipelines |
| Preflight | The check before DDL runs that stops changes which would break dependents |
| Promise | An explicit check on a dataset, evaluated on a schedule. See Datasets & promises |
| Query shape | Queries that are the same except for literals (normalized_query_hash), tracked together for performance |
| RBAC database | CHouse UI's own SQLite or PostgreSQL database for users, settings and evidence |
| Remediation credential | A separate ClickHouse login per connection, the only one fixes run with |
| Role | A bundle of permissions and data access policies. Six are built in. See Users & roles |
| Root cause chain | The computed path from an incident's symptom down through data, transform, ingestion, external and engine layers |
| Scheduled query | A SQL job on a schedule or after another job. See Scheduled queries |
| Trust state | A table's state from its baseline: learning, trusted, degraded or stale |
| Watcher | A plain-language sentence that Chouse AI turns into a promise for you to review. See Context |