Glossary

Plain definitions for the terms Strata uses.

Agent

A scheduled, autonomous run of Strata that does work without a person present. Agents are created from the Agents page in the sidebar, run on a cron-style schedule, and can optionally send a notification email when a run finishes. An agent can also be given its own webhook URL so an outside system can trigger a run — agent webhooks are inbound triggers only; Strata never calls out to a webhook of yours. An agent's tool access is narrower than a chat conversation's: capabilities are granted one at a time from a fixed list — SQL, Outlook read, Outlook send, OneDrive read, OneDrive save, Excel, PDF, Python, invoking another agent, web search, Jira read and write, Confluence read and write — and the server filters every run's toolset down to those grants, further intersected with what the organization, the role, and the owner's per-user extension mask allow.

AI Engine

The Azure AI Foundry instance Strata routes every AI request through. Every organization connects its own Azure AI Foundry resource (bring your own / BYO), configured from Admin -> Organization Settings -> AI Engine. All chat and every scheduled agent run goes through your Foundry, so inference stays inside your own Azure tenant.

API

Strata's HTTP API — a server-to-server interface for running prompts, executing SQL, generating documents, and managing conversations from your own backend. Every request authenticates with an API key over Authorization: Bearer; there are no cookies or browser sessions, and endpoints live under /v1/. See the API overview.

API key

A long-lived, server-only secret (prefixed sk_strata_live_) that authenticates HTTP API requests. Admins mint, scope, and revoke keys from Admin -> API Keys; the full secret is shown only once, at creation. Each key carries its own scopes, a sliding-window rate limit, and an optional monthly budget. See Authentication.

Artifact

A persistent side-panel resource tied to a conversation. There are two artifact types today:

SQL workspace — opens when the AI writes or modifies a SQL query. The latest edit is shown, runnable, and survives page reloads.
Dashboard — opens when the AI builds an interactive multi-component dashboard via update_dashboard.

Artifacts persist across page refreshes and conversations (server-side, on the artifacts table) and are cleaned up when their parent conversation is deleted.

Audit log

The tamper-evident record of every administrative action: who did what, to whom, with what details, and when. Configured retention (30 to 3650 days) is set at Admin -> Organization Settings. Viewable from Organization settings -> Audit if your role grants canViewAudit.

Callout

A styled card the AI emits in chat to highlight a fact, warning, success, or other signal. Eight types are available — see Callout reference.

CSRF token

A short cookie value (strata_csrf) that the Strata frontend automatically attaches as the X-CSRF-Token header on every mutating API call. The server rejects mutating requests whose header does not match the cookie. You never set this manually.

Connection (database connection)

A saved configuration for talking to a SQL Server database — host, port, database name, credentials. Strata supports multiple saved connections per user; one is "active" at a time. Connection credentials are encrypted with AES-256-GCM. Organizations can also publish org-managed connections that admins maintain centrally and assign to users or roles.

Confluence

Strata's extension for Atlassian Confluence. It shares one Atlassian connection with Jira — a single Atlassian card in Settings → Extensions, one consent, one pinned site — but is governed as its own extension, with its own availability switch on the Atlassian extension page (Settings → Extensions → Atlassian → Manage), its own writes toggle, its own agent capabilities, and its own API scopes, so an admin can allow one product and deny the other. Strata searches pages, reads a page as plain text (macros are not expanded and the page author is not available), and lists spaces. An optional space allow-list narrows what it can reach — space keys are case-sensitive here, unlike Jira project keys — and it fails closed. Writes are off by default and resolve through the same three layers as Jira's, independently of them; when permitted, Strata creates pages and adds comments. There is no tool that edits an existing page. Strata never indexes Confluence into the knowledge base. See Atlassian extensions.

Extension

An external capability Strata can use. Current extensions: SQL Server, OneDrive, Outlook, SharePoint, Microsoft Teams, Web search, Jira, and Confluence — the last two being the ones that connect to something other than Microsoft or your own database, and the two that share a single Atlassian connection while staying separately governed. Enabling an extension expands the AI's tool set with that service's capabilities. Admins choose which extensions are even available org-wide from each extension's own page — Settings → Extensions → (extension) → Manage → Organization settings — and a role can narrow that further per member.

In-app support

The help assistant in the bottom-right corner of the Strata app — an AI agent that answers product questions instantly, alongside quick-answer FAQ articles and a link to email the human support team. See Getting help in Strata.

Inbox

Your in-app message center, opened from the Inbox tab in the sidebar. It collects agent run notifications, organization announcements, system notices, and the new-user welcome message. The sidebar shows unread items; the full archive lives at /inbox/all. See Inbox.

Incognito (chat)

A per-conversation toggle in the chat composer that suspends all persistence for that conversation: no messages, AI output, memory updates, or voice transcripts are stored. Token usage is still billed and counted toward your quota, and the admin Usage page still sees aggregate spend, but not message content.

Jira

Strata's extension for Atlassian Jira, covering both Jira Software projects and Jira Service Management (JSM) service desks. It shares one Atlassian connection with Confluence but is governed as its own extension. An admin picks one of two connection modes from Settings → Extensions → Atlassian → Manage, and the choice applies to both products: each person signs in, where every member connects their own Atlassian account and their own Jira permissions bound every result, or one organization account, where an admin configures a single Atlassian service account everyone works through. The Atlassian site is pinned per organization and re-checked on every request, and an optional project allow-list narrows search, counts, single-issue reads, and JSM queues further. Writes — creating an issue, adding a comment — are off by default. Turning them on is an organization decision that a role and an individual member can each narrow, never widen: whether one person may write is the intersection of the three layers, and it fails closed. When a write is permitted, chat asks you to approve it, it carries a trailing line naming Strata and is recorded in the audit log, and comments on JSM requests are always posted as internal notes the customer who raised the request never sees. Strata never indexes Jira into the knowledge base. See Atlassian extensions.

Jira Service Management (JSM)

Atlassian's service-desk product, reached through the same Jira extension as Jira Software. Strata can read service requests and list queues, and — only when Jira writes are permitted for the person the run belongs to and the separate organization setting for commenting on service requests is on — add comments, which are always internal notes and never visible to the customer who raised the request. A member denied Jira writes cannot post an internal note either.

Knowledge base

A curated index of your organization's SharePoint documents that Strata searches to ground its answers with cited sources (retrieval-augmented generation). Admins curate sources from Admin → Organization Settings → Knowledge Base (org setup requires the OneDrive extension); a member needs the SharePoint extension in their role to trigger KB search in chat. See Knowledge base.

Legacy model

A catalog model marked legacy — an older model kept available for a transition period after a newer one supersedes it, so members can migrate chats, agents, and saved workflows before it is retired. Legacy models are flagged in the picker and discouraged for new work. Admins choose whether to keep a legacy model enabled in the Model Catalog; when a model is fully retired, Strata can auto-migrate affected work to its named replacement. See Choosing an AI model.

Memory

The AI's mechanism for remembering facts across conversations. Two scopes:

User memory — personal facts about you, available in every chat.
Per-connection memory — facts about a specific database, available only when that database is the active connection.

Memory is managed silently via the memory_update block; you can view, import, delete individual entries, and clear all memory from Settings -> Capabilities.

Mode

A conversation-level setting that controls how technical the AI's answers are:

Auto — adapts to your apparent expertise dynamically. Default.
Technical — shows every SQL query, uses precise database terminology.
Simple — hides SQL entirely and explains results in plain language.

Model access (per-user)

A per-user restriction on which AI models a single member can use, set from Admin → Users → (manage user) → Extensions & Data → AI models. Leave every box checked to inherit the full set the organization has enabled; uncheck models to narrow that member further. The member's effective models are the intersection of the organization's enabled models, their role's allowed models, and this per-user list — and that intersection is enforced across chat, scheduled agents, and the Teams bot. See Choosing an AI model.

Model Catalog

The curated set of AI models available in Strata. Kronisys maintains the catalog centrally — provisioning each model in Microsoft Azure AI Foundry and publishing it with its capabilities, context window, per-1M-token pricing, and Strata Benchmarks. Each organization enables the subset it wants from Admin → Organization Settings → Models → Model Catalog; only enabled models appear to members. The catalog spans many models across OpenAI, Anthropic, xAI, DeepSeek, Mistral AI, Cohere, and Microsoft, plus specialized voice, transcription, embedding, and rerank models. Open the catalog in-app for the live list rather than relying on a fixed count. See Choosing an AI model.

Organization

A Strata tenant. Every user, conversation, connection, setting, audit-log entry, and artifact lives inside exactly one organization. Organizations are fully isolated from each other — no data ever crosses tenant boundaries.

Organization announcement

A message an admin broadcasts to members' inboxes — targeted to the whole org, specific roles, or named people — with optional priority, a call-to-action button, and read receipts only the author can see. Composed from Admin → Announcements. See Announcements.

Permission tier (SQL)

The per-user SQL Server access level:

Viewer — read-only. Cannot modify data.
Editor — INSERT, UPDATE, DELETE allowed, but every write requires per-query approval.
Admin — full SQL including DDL (CREATE, ALTER, DROP). Reserve for DBAs.

The default for new members is set during onboarding and editable from Admin -> Organization Settings.

Role

A bundle of granular permissions assigned to a user. Built-in roles are User (standard access) and Admin (can manage org settings). Admins can create custom roles from Admin -> Roles. Permissions include canAccessAdmin, canManageUsers, canManageRoles, canRevokeSessions, canInviteUsers, canManageOrgSettings, canManageOrgConnections, canViewAudit, and others.

SCIM

System for Cross-domain Identity Management — the standard for letting an identity provider create, update, and deactivate users in Strata automatically. Strata exposes SCIM 2.0 endpoints at /scim/v2/ (Users, Schemas, ServiceProviderConfig) authenticated by a per-organization bearer token. Tokens are issued and revoked through the admin API (/api/admin/scim/tokens) — see SCIM provisioning.

Session

An authenticated browser session. Lives in two places: in-memory (fast lookup) and the sessions table in the platform database (survives server restarts). Sessions expire after 12 hours of inactivity; active sessions are extended on each request up to a 7-day absolute cap from the original login. Admins can revoke any session from Admin -> Active Sessions; the next API call on a revoked session returns 401 and returns the user to the sign-in screen.

Skills

Specialized, on-demand knowledge bundles the AI can pull in to handle particular tasks (Excel generation, PDF reports, OneDrive workflows, dashboard design, and others). Browse the Skills directory from Settings -> Extensions (the Skills Library panel) or the composer's plus menu, and pin as many as you like so the AI does not have to search for them first.

Spend ceiling

An organization-wide monthly limit on total AI spend. When the running total reaches the ceiling, Strata pauses AI across every surface — chat, scheduled agents, the Teams bot, the API, and voice — cutting off mid-response if a request is in flight, so the member sees a "response was cut short" notice rather than an unbounded bill. Reaching the ceiling posts a notice to the admin inbox; the limit resets at the start of each month, and leaving it blank means no cap. Only owners whose role grants canManageBilling can set it, from Admin → Billing.

Strata for Outlook

The Strata task-pane add-in that runs inside Outlook (desktop and Outlook on the web). It reads the email you have open and its full thread so you can summarize it, ask questions about it, and draft replies without leaving Outlook. This is distinct from the Outlook extension, which lets the chat AI search, read, and send mail through tools. The add-in is installed by sideloading its manifest or deploying it org-wide through Microsoft 365 integrated apps. See Strata for Outlook.

Suggestions

The three follow-up question cards at the end of most data-aware AI responses. The AI is instructed to derive them from the actual data shape returned. Click one to send it as your next message.