Glossary

This glossary is the single source of truth for FlowLakes terminology. Each concept has one name, used consistently everywhere. Where two ideas could collide, the entry says so.

Platform data

Table. A tenant-scoped set of records with a typed schema. See Tables and records.

Record. One row on a table. For a child table, a record spans the child and its ancestors under a shared id.

Inheritance chain. The ordered ancestry of a table, from the child up to the root. A child holds its own columns, and its ancestors hold the shared ones.

fl_tasks. The shared work-item table and the common parent. It holds state, priority, assignment, numbering, the activity timeline, and SLA due windows.

Task. A record on a table whose inheritance chain includes fl_tasks.

Configuration item (CI). A canonical asset in the CMDB, stored in fl_cmdb_items and linked by typed edges. See CMDB and discovery.

Automation

Business Rule. Declarative trigger-condition-action logic bound to a table that runs on record events. Use this term rather than “automation rule.” See Business rules.

Workflow. An asynchronous, multi-step process triggered by an event, a schedule, a manual run, or another workflow. See Workflows.

Service management

Service Offering. The routing spine for intake. A published offering stamps the assignment group and SLA on a new task. See Service Hub.

Catalog item. An intake template, a form plus a target table, in Service Hub.

Approval. A gated decision on a record. An Approval Policy self-triggers on its target tables and routes the decision. See Approvals.

Notification Rule. Configuration that fires on an event, resolves recipients, renders a template, and queues delivery. See Notifications.

SLA. A commitment to response and resolution windows, stamped on the task and measured in working time. See SLA.

Home (application). The page an application opens on: attention cards with live counts, quick actions and the application’s dashboard charts. Each card counts the same filter as a list view, so the two always agree. See Work incidents.

List view. One named way of looking at a list, shown as a tab with its count above the list, such as Open or My Work on Incidents. A filter you add narrows the tab you are on. See Work incidents.

Setup hub. An application’s Configuration hub: one card per thing the application needs, each with a status line, a state and one action. See Configure an application.

License Metric. A definition of how a software license is counted, used to compare consumption against entitlement. See CMDB and discovery.

Data Source. A configured connection that brings external data into the platform, either as records in a table or as discovered assets. See Data sources.

Agentic

Agent. A tenant-scoped AI identity with a service account, a charter, granted tools, and an autonomy mode. See Agents.

Service account. The user identity an agent executes as. It is the permission boundary an agent’s granted tools are checked against.

Autonomy mode. One of propose, bounded, autonomous, or investigate. It decides how an agent’s writes are gated.

Records Reader. The built-in role (agent_record_reader) that lets an agent’s service account read your tenant’s records generally instead of a named list of tables. Credentials, platform configuration, the access model, audit logs and skill bodies stay closed. See Safety model.

Finding. The structured result an investigate-mode agent ends its run with: a summary, the evidence it rests on, an optional recommendation, and an optional suggested plan as plain data. See Agents.

Skill. A versioned prompt fragment with activation rules, stored in fl_skills. An agent’s charter is a Skill. See Skills and knowledge.

Knowledge Plane. The part of the platform that assembles every system prompt from active skills.

Draft skill. A captured skill with draft status. It stays out of every prompt until a human activates it.

Plan. An agent’s typed, approvable proposal on a record. See Plans and governance.

Judgment. A single-shot AI evaluation of one record with a typed output contract, a confidence threshold, and a governed apply policy. Shown as Decisions in the admin UI (AI → Decisions); the API and capability names keep judgment. See Judgments.

Decision (judgment). The audit row one judgment evaluation leaves: input snapshot, output, confidence, rationale, and its outcome, including how a human resolved it.

Review queue. The set of pending judgment decisions awaiting a person, either routed there by policy or held back by the confidence threshold. See Judgments.

Judgment version. One immutable snapshot of a judgment’s configuration. Direct saves record versions automatically; a draft version activates only through the eval gate. See Judgments.

Eval case. A labeled example for one judgment: an input snapshot plus the expected output. Captured automatically from review-queue approvals and overrides, or authored by hand.

Eval gate. The rule that a draft judgment version with at least one enabled eval case cannot activate until an eval run has scored that exact draft. The gate guarantees the eval happened; the human judges the score.

Escalation investigator. The investigate-mode agent a judgment names for deeper review. A reviewer dispatches it from the review queue and reads its finding beside the pending decision. See Judgments.

Your instance

Instance. One customer’s FlowLakes environment at its own address, such as acme.flowlakes.com. Each instance is one tenant. See Tiers.

Tier. Where an instance runs: the shared trial, a dedicated database, or a dedicated stack. See Tiers.

Subscription plan. The commercial plan an instance is on (trial, team, business or enterprise), which sets its usage limits. Not to be confused with a Plan, which is an agent’s proposal on a record. See Plans and the trial.

Trial. The subscription plan a self-serve instance starts on. It runs for 30 days from creation. See Plans and the trial.

Suspended instance. An instance that is switched off, for example after its trial ended. Nobody can sign in and nothing runs, but its data is kept for a published period. See Plans and the trial.

Access and secrets

Capability. A platform permission named platform.<area>.<read|write> that governs the tables its area owns. See Identity and access.

External Capability. Outbound reach for an agent, such as web fetch, email, Graph, or REST. A Capability is an in-platform permission, while an External Capability is outbound reach. See External capabilities.

RBAC. Per-table read, create, update, and delete authorization for application tables that no capability area governs.

Credential. A typed, encrypted secret in fl_credentials. See Credentials and secrets.

Tenant. An isolated customer boundary. Every query, cache key, and audit entry is tenant-scoped.