Skip to Content
Baseera Docs

Glossary

Short definitions of the terms used across the Baseera docs.

Agent

One assistant, with a system prompt, a model, a tone, knowledge and a theme. It has a draft and published versions.

Approval

The user's decision on a draft or write tool call, made on a card in the chat. Only an approved call runs.

Allowed domains

The list of domains that may start a session with an agent's publishable key. Requests from other origins are refused.

BYOK (bring your own key)

Baseera runs on your model provider key. You pay the provider directly.

Check

In evals, a typed test on a run, such as a tool being called or the reply language. In immersive mode, a rule about a record that can raise a flag.

Citation

A marker such as [S1] in an answer that points to a source passage. Sources are listed with the message.

Draft (tool kind)

A tool kind that proposes a change and waits for approval, like a write tool.

Draft (version)

Unpublished edits to an agent. Production does not change until you publish.

Eval

A test of the agent: a case, its mocks and its checks, run on a draft or a published version.

Immersive mode

A mode where the agent works on the one record the user is editing, flags problems and proposes fixes.

Judge

A model that grades a reply against a rubric when a rule cannot.

Knowledge base

A set of documents that agents search and cite. It has its own embedding model and provider key.

Mock

A fixed or recorded tool result used when a test replays a conversation, so the test needs no live app.

Organization

Your workspace in the dashboard. It owns agents, keys, knowledge, members and the audit log.

Production

The published version of an agent that the runtime serves.

Provider key

Your key for a model provider, stored encrypted by Baseera and never shown again.

Publishable key

A pk_ key that identifies an agent in the browser. Safe to expose, and limited to allowed domains.

Runtime

The API that runs a chat turn: it checks credentials, builds the prompt, searches knowledge, calls the model and streams the reply.

Secret key

An sk_ key for your server that signs user tokens and server calls. Shown once.

Session

One conversation between one user and one agent.

Tenant

One of your customers, as your app knows them. It comes from the user token and scopes knowledge and sessions.

Tool

A function in your app that an agent may call. It has a kind: read, draft or write.

Turn

One user message and the agent's full response, including any tool calls and their results.

User token

A short-lived signed token from your server that says who the user is and which tenant they belong to.

Version

An immutable published snapshot of an agent's configuration. You can roll back to an earlier one.

Next steps

On this page