Glossary

Short, precise definitions of the terms used in the Imaginne documentation, in alphabetical order. Each term points to the page where the topic is covered in depth.

A

Agent — The entity that drives a session: it reads, reasons with a model, uses tools and skills, and responds. On Desktop/TUI/VS Code the agent runs in the local Engine. See Architecture.

AllowedModels — The list of models the organization's policy permits. It determines what appears in the model picker. See Models and Governance.

Artifact — A file the agent generates: .pptx, .docx, .xlsx, .pdf, images, code. On Desktop it appears in the Preview dock and in the output folder; in web chat, in cards with Download. See Files & artifacts.

Automatic mode — The per-turn routing between nnumbers and nnumbers-code according to the task. See Models.

Autonomy — How much the agent asks before acting: low (confirms everything), medium (default), and high (does not ask). It is distinct from the execution mode. See Autonomy & permissions.

B

BYOK (Bring Your Own Key) — Model-vendor keys registered by the organization (not per user), write-only, used server-side when calling the vendor. See Administration → BYOK and Credentials & secrets.

C

Command vault (firewall) — The always-on set of rules that confines the agent to the workspace and imposes fixed denials (catastrophic commands, system paths). It applies at any autonomy level. See Autonomy & permissions.

Compaction — The automatic reduction of older parts of the context when the window approaches its limit, so the conversation can continue without losing the thread. See Sessions.

Content checker — The layer that redacts known sensitive patterns (emails, national IDs, cards, tokens, keys) before sending the prompt to the model, without logging the text. See Local execution.

Context — What the model "reads" in a call: the current message plus the history and the relevant excerpts. It has a limit (the context window). See Sessions.

D

dangerous_bypass — The least restrictive execution mode, which turns off shell validations. It is only available if the app was started with the appropriate flag; even then, workspace and auditing still apply. See Autonomy & permissions.

Desktop — Imaginne's native app (macOS/Windows), with the Engine embedded. See Install Desktop and Use Desktop.

E

Engine — The component that runs the agent. It runs locally on Desktop, in the TUI, and in VS Code; in web chat it runs on the server. See Architecture.

Env-secret — A per-organization secret (token, certificate, credential), write-only, injected only into the skills that declare it in required_env, with no carry-over. See Administration → Env-secrets and Credentials & secrets.

Execution mode — The control, separate from autonomy, over running terminal commands: safe, auto (default), and dangerous_bypass. See Autonomy & permissions.

F

Firewall — See Command vault.

G

Gateway — Imaginne's model broker: the component that talks to the AI provider and the single egress point for the prompt content. See Architecture.

I

Identity — Who you are in Imaginne, carried by your authenticated session and tied to your organization. See Identity & login.

L

local_plain — The normal skill execution mode: the files sit open at ~/.imaginne/skills/<key>/, transparent. See Execution modes.

local_protected — The execution mode with verified integrity (HMAC + digest), execution in a temporary folder, and anti-shadowing. It guarantees integrity, not secrecy (it is not DRM). See Protected skills.

Local-first — The principle that, on the app surfaces, the agent runs on your machine and files do not leave; only the prompt content travels. See Local execution & privacy.

M

Memory — Four Markdown notes per project (profile, project, glossary, decisions) that persist across sessions and enter the context at the start of each one. See Memory.

Model — The AI engine the agent uses. The product exposes nnumbers and nnumbers-code, plus the automatic mode. See Models.

N

nnumbers — The product's main and default model: it plans, reasons, and acts. See Models.

nnumbers-code — The model specialized in development: code, diffs, scripts, tests, and debugging. See Models.

O

Organization — Imaginne's tenant. It defines policy, credentials, users, and skills; everything you do happens in its context. See Organizations & policies.

outputs/ — The project's default output folder, where skills write artifacts. See Files & artifacts.

P

Platform — The control plane: identity, policy, remote relay, and the /app console. It does not talk to the model provider (that is the Gateway's job). See Architecture.

Policy — The set of organization rules that governs models, credentials, skills, execution, and content. It combines layers with deny-wins. See Organizations & policies and Governance.

Profile — The entity that links skills to users. Without a profile that has skills, the agent receives no capabilities for you. See Administration → Profiles.

R

Remote (pairing) — Driving your local session from another screen. Your app is the runtime; Platform only does the relay. Pairing via a 30-second ephemeral code. See Use remotely.

required_env — The list of env-secret names a skill declares it needs. The Engine injects only those names into the skill's process. See Skill env-secrets.

Routine — A recurring task, user-global, that runs while Desktop is open (a real scheduler). Distinct from Saved Prompts. See Routines.

S

safe — The most restrictive execution mode: no shell strings, operators, or redirection. See Autonomy & permissions.

Saved Prompt — A reusable prompt, per project and manual only ("Run Now"), with no scheduling. Distinct from Routine. See Routines.

Session — A continuous conversation with the agent: history, actions, and state. You open, close, and resume it. See Sessions.

Skill — A capability the agent uses when relevant. It is not a command (/x) nor a file mention (@x). Sources: bundled, published by the org, personal/project. See Skills — overview.

Skill group — A grouping of skills that also delegates administration to a skill_admin. See Administration → Skill groups.

Skill Studio — The conversational skill editor and generator in the /app console. See Create a skill.

skill_admin — The role that manages the skills of the groups under its administration. See Administration → Users & roles.

Streaming — The incremental delivery of the model's response, token by token, as it is generated. See Sessions.

T

Token — In the identity sense, see Identity; in the sense of a model's unit of text, see the context window in Sessions.

TUI — The terminal interface (the imaginne binary), with the local Engine. See Install the TUI and Use the TUI.

V

VS Code (extension) — The nnumbers.imaginne extension, which uses the imaginne serve CLI underneath. See Install in VS Code and Use in VS Code.

W

Workspace — The project folder that delimits where the agent can read, write, and execute by default. The command vault imposes that limit. See Autonomy & permissions.