Glossary
Every SimplAI-specific term with a short definition and a link to the deeper docs where applicable.
A
Agent — A first-class application type for goal-driven, autonomous behaviour. See Agents.
AI Chain — The legacy term for a chat-first workflow. Now unified into Tools with Conversation mode.
Application — Any deployable unit: a Tool, an Agent, or a Voice Agent. Applications have versions, can be exported, and are surfaced via the Deployment tab.
ASR — Automatic Speech Recognition. The Voice Agent transcriber step. See Voice Agent transcriber.
Async (execution mode) — Voice Agent mode that routes calls through the dedicated voice priority queue. See Execution modes.
Audit log — Account-wide log of CRUD actions on every resource. See Audit logs.
Autofill with AI — Generates an evaluator's prompt or code from its name and description. See Autofill.
B
Base instruction — The system prompt for an Agent or Voice Agent.
BYOK (Bring Your Own Key) — Use your own provider API key for LLM, STT, TTS, or evaluator providers. See Voice Agent BYOK.
Bulk Run — Execute a workflow across a list of inputs in parallel with configurable concurrency. See Bulk & Scheduled Execution.
C
Cloner — In the Marketplace context, a user who clones a shared agent into their project. See Cloning.
Conversation mode — A toggle on every Tool that switches it from typed inputs + JSON outputs (default) to Chat History input + single reply output. Replaces the standalone AI Chains primitive.
Custom Attributes — Per-application secrets / constants accessible via {{custom_attribute_name}}. See Studio canvas.
D
Deployment tab — The per-application surface for exposing the app (API, Embed, Webhooks, A2A, channel integrations). Formerly named "Integration".
E
Embed — A copy-paste widget that drops a SimplAI agent into a web page.
Endpointing — Voice Agent STT setting: silence (ms) before speech is considered complete. See Transcriber.
Evaluator — A scoring unit applied to a test run. Two types: LLM-as-a-Judge and Python Executor. See Evaluators.
F
Fail Branch — A step's error handling mode where the step produces two outputs (success / failure). See Error handling.
Fallback Value — A step's error handling mode where the step's outputs are set to user-defined values on failure.
G
Guardrails — Validators that check input or output for safety / compliance. See Guardrails.
J
Jinja templating — The {{ }} syntax with conditionals and filters, evaluated at execution time. See Studio canvas.
K
Karix — Third-party RCM provider for WhatsApp. See Karix WhatsApp.
Keyterms — Domain-specific terms passed to a STT model to improve recognition (Deepgram Nova-3 only).
KB (Knowledge Base) — Searchable, AI-ready store of indexed documents. See Knowledge Bases.
L
LLM-as-a-Judge — An evaluator type that uses an LLM with a Jinja prompt to score outputs.
M
Marketplace — The Internal Marketplace where Builders share agents within their tenant and Cloners discover them. See Templates & Marketplace.
MCP — Model Context Protocol. Tool integration standard SimplAI supports as both server (host) and client.
Multilingual ASR/TTS — Voice Agent option for multi-language transcription and synthesis. See Multilingual.
N
Named Guardrails config — A reusable bundle of guardrails defined at the project level. See Project-level policies.
O
Observability — The visibility layer of the platform — live tracing, run history, analytics, audit logs. Formerly named "Tracing". See Observability.
P
Planning Agent — An Agent with the planning capability enabled — includes a mid-execution evaluator and next_goal narration.
Project — The unit of grouping inside a workspace. Users have project-scoped roles; resources live within a project.
Published version — A version of an Application that is callable via the Deployment tab. Distinct from a Draft (editable, not callable).
R
Realtime API (execution mode) — Voice Agent mode that runs through the OpenAI Realtime API instead of the standard STT → LLM → TTS pipeline. See Execution modes.
Run — One execution instance of an Application.
S
Setup Incomplete — A badge on a cloned agent that's missing credentials from the setup dialog.
Skills — A capability of Agents that brings harness orchestration mode. See Agents.
STT — Speech to text. See ASR.
Sync (execution mode) — Voice Agent mode that bypasses queues for low latency. See Execution modes.
Sync mode (Tools) — A Tool-level toggle that executes steps immediately without queueing. Required for tools attached to Voice Agents.
T
TTFS (Time to First Speech) — Time from user speech-end to agent audio-start in a voice run. The headline voice latency metric. See Voice latency.
TTS — Text to speech.
Trial session — Sandboxed Marketplace trial where a Cloner tests an agent before cloning.
V
VAD (Voice Activity Detection) — Detection of when a user has started / stopped speaking. See VAD.
Voice Agent — A dedicated voice-first application type. See Voice Agents.
Voice Agent node — A workflow node that invokes a Voice Agent (typically outbound calling) from inside another Tool. See Voice Agent node.
Voice cloning — Use a custom voice ID for the TTS step (ElevenLabs-specific in M2). See Voice (TTS).
Version diff — Side-by-side comparison of two versions of an Application's configuration. See Versioning.