Terms and definitions

Use this glossary to look up the terms used throughout Orek’s documentation. For your first installation, start with Quickstart.

Providers and models

Term	Meaning
Provider	Orek’s connection to a service that runs models. A provider can connect to a cloud service or a local server.
Model	The AI used for a particular job, such as generating a response, creating embeddings, or ranking search results.
Model catalog	The model definitions available to Orek, including built-in entries and models you add in configuration.
Model selector	The name used to choose a model from the catalog, such as openai/gpt-6-luna.
Model pin	An explicit model choice used instead of automatic routing. A session selection is temporary; a saved pin persists in configuration.
Provider pin	A restriction that keeps model selection within one provider. It does not identify a particular model.
Route / model routing	A route describes a kind of task and model preferences. Routing uses those rules and the request to select an eligible model when no model is pinned.
Cloud model	A model running on a remote provider’s infrastructure. Internet access is required to use a cloud-hosted provider.
Local model / local inference	A model running on your own machine. Inference is the act of running a model to produce a response, embedding, or score.
Local model server	A process, such as llama.cpp, Ollama, or LM Studio, that exposes local models through an API Orek can call.
Endpoint / base URL	The network address Orek uses to reach a provider or model server.
Provider API key	A credential used to authenticate with a model provider. It is separate from your Orek license.
OS credential store	Your operating system’s own storage for secrets, also called the keyring: macOS Keychain, Windows Credential Manager, or Linux Secret Service. Orek keeps API keys you enter during setup there by default (the keyring storage option), out of your configuration files.
BYOK	Bring your own key: connect Orek to your own model providers and pay those providers directly for usage.
Reasoning replay	What happens to a model’s earlier reasoning when you switch models. drop (the default) sends none from before the switch, except to Kimi Code, which requires its own; preserve sends each provider its own. Set it with agent.reasoning_replay or --reasoning-replay.
Segment	The stretch of a conversation handled by one model. Each model switch starts a new segment, and /usage reports cost per segment.

See Providers, Model routing, and Local models for setup and selection.

Working with Orek

Term	Meaning
Agent	The model working with Orek’s instructions, context, and tools to carry out a task.
Agent loop	How Orek carries out a turn: it sends requests to the model, runs the tool calls you permit, returns the results to the model, and saves the conversation. A turn can contain several model requests and tool calls.
Project / workspace	The folder Orek is working in, including your source files and project-specific .orek/ state.
Session	A conversation and its saved state. You can continue the same session across multiple turns or resume it later.
Turn	One execution of a request within a session, including the model responses and tool calls needed to finish it.
Prompt	The request or instructions you submit to Orek.
Composer	The input area where you write prompts and commands in Orek’s terminal interface.
TUI	Terminal user interface: the interactive Orek application you open by running orek.
CLI	Command-line interface: commands run from your shell, such as orek config or orek run.
Headless run	An Orek execution without the interactive terminal interface, typically started with orek run for scripts or automation.
Slash command	An Orek interface command beginning with /, such as /model or /context. It is entered inside Orek rather than in your shell.
Tool / tool call	A tool is an operation available to the agent, such as reading a file or running a command. A tool call is a request to execute that operation with specific arguments.
Tool discovery	Finding available tool definitions when the agent needs them. Discovery does not grant additional permission to execute a tool.
Provider web tools	Web search and page fetching that the model provider runs on its own servers, such as OpenAI’s web_search. Orek’s own webfetch tool fetches pages from your machine instead.

See Using Orek, Manage sessions, and Agent tools.

Context and code retrieval

Term	Meaning
Context	The information included in a model request: instructions, conversation history, tool definitions and results, and relevant project content.
Context reference / @ reference	An explicit file, line range, or symbol you identify in a prompt so Orek can include it as context.
Token	A unit a model uses to represent text, often part of a word. Token counts measure request and response size, and most providers bill by them.
Context window	The model’s capacity for input and output, measured in tokens.
Input window	The portion of the context window available for a request after reserving the model’s output limit.
Token budget	A limit on how many tokens a particular part of the work may use, such as retrieved context or a model response.
Compaction	Summarizing earlier conversation content to reduce what the model receives on later requests. It does not roll back project files.
Memory	Facts saved for use across sessions, separately from an individual conversation’s history.
Chunk	A portion of project text stored and searched as one retrieval unit.
Embedding	A numeric representation of text used to find content with similar meaning. An embedding model produces those representations.
Index	Searchable data about your project’s files that Orek builds and can rebuild at any time. Orek stores it in .orek/index/.
Hybrid retrieval	Finding relevant project text by combining meaning-based search, full-text search, and exact matches, then choosing the best results that fit the token budget.
Automatic retrieval	Orek finding and adding relevant indexed project context when preparing your request. Explicit @ references let you choose context directly.
Reranking	An optional additional scoring step that reorders retrieved candidates for relevance to a query. A reranker is the model used for that step.
Symbol	A named code element, such as a function, class, or type.
Code graph	The symbols in your code and how they relate, such as definitions, callers, and imports. It is part of the project index but doesn’t use embeddings.

See Add context, Manage context, Find and explore code, and Code graph.

Plans, workflows, and delegation

Term	Meaning
Plan / Plan mode	A plan describes a proposed approach. Plan mode lets the agent investigate and prepare that approach while implementation actions are blocked. Approving a plan starts implementation under your existing permission rules.
Skill	Reusable instructions for a task, loaded through a $name mention. A skill does not grant additional tool permissions.
Workflow / template	A reusable definition of steps and their dependencies. Orek decides when each step runs; models do the work within each step. The TUI also calls a template a recipe.
Workflow run	One execution of a workflow template with a particular input and recorded step results.
Workflow step / dependency	A step is one unit of work in a template. A dependency names another step that must finish first; independent steps can run in parallel.
Subagent	A delegated agent that performs a focused part of the work with its own conversation, within the parent session.
Subagent profile	Named settings for a subagent’s role, instructions, tool list, and optional model choice, such as the built-in explore or edit. A profile’s tool list focuses its work; session permissions still apply.
Goal	The overall objective recorded for a piece of work.
Task / checklist item	An entry in a saved, ordered checklist, with its own status. A subagent can be linked to one through work_item_id. A checklist tracks progress; it is separate from a workflow and its steps.
Auto-continue	Continuing work on unfinished checklist items after a turn ends, when auto-continue is on (agent.auto_continue or /auto-continue).

See Plan and track work, Skills, Workflows, and Subagents.

Permissions and execution

Term	Meaning
Permission rule / policy	A rule that restricts or permits an action, or requires approval. Tool availability alone does not authorize an action.
Approval mode	The setting that controls when Orek asks for approval before executing actions. Permission rules still restrict what can run.
Auto-review (auto_review)	An approval mode that uses local risk scoring for eligible shell commands. Commands that need human review are sent to the approval prompt.
Shieldstral	The local model Orek uses to assess shell-command risk in auto-review mode.
OS sandbox	Optional operating-system restrictions applied to execution. The available containment and environment filtering depend on the platform.
Execution policy	An experimental setting (exec.default_sandbox) that decides whether approved actions run (approved, the default), return a preview (dry_run), or are only described (simulation).
URL allowlist	A configured set of allowed network destinations. Orek uses it to restrict URL access; configuring one turns off provider web tools.

See Permissions and security.

Configuration and integrations

Term	Meaning
Configuration key	The name of a saved setting, such as ui.theme. Its documented scope determines where it can be set.
User configuration	Your personal Orek settings, stored outside the project and reused across projects.
Project configuration	Settings stored in the project’s .orek/config.toml. Only settings that support project scope can be changed there.
Session override	A temporary choice that applies to the current session without changing your saved configuration.
TOML	The text format used by Orek’s configuration files. It groups named settings into tables such as [index].
Environment variable	A named value supplied by the process environment. Configuration can reference one, for example ${OPENAI_API_KEY}, without saving its secret value in the file.
MCP	Model Context Protocol: connects Orek to external servers that expose tools, prompts, and resources.
ACP	Agent Client Protocol: connects an editor to Orek as its coding agent.
LSP	Language Server Protocol: connects Orek to language servers for code diagnostics such as errors and warnings.

See Configuration, MCP servers, IDE integration, and Language servers.

Licensing and releases

Term	Meaning
Console	The Orek website area where you access your licenses and manage approved devices.
License / license key	Your entitlement to use Orek and the key used to activate a paid license. This does not provide a model-provider account or pay for model usage.
Activation	Authorizing a device to use Orek under your trial or paid license.
Perpetual license	A license that lets you keep using the purchased major release series without a subscription.
Major release / included updates	The release series covered by a license, such as 2026.x. Minor and patch updates within that purchased series are included.
Release archive	The downloadable package containing Orek and its bundled runtime for a particular platform.

See Activate your license, System requirements & release status, and Upgrade Orek.