31 concepts. Plainly explained.

Personal AI assistant

An AI that works for one person — handling their inbox, calendar, channels, and recurring tasks across time.

Multi-channel AI

AI assistants that work across messaging surfaces — web, WhatsApp, Telegram, Discord, voice.

Always-on agent

An AI agent that runs 24/7 — handling scheduled tasks, watching for events, never sleeping.

Proactive AI

An AI that initiates — pings, suggests, reminds — instead of waiting for prompts.

Self-improving AI

An AI that learns from your interactions and improves how she serves you over time.

Personal CFO

An AI that handles money like a CFO would — tracking, anomaly detection, tax prep.

Chief of staff

A senior generalist who runs the operational side for an executive — calendar, prep, decisions, follow-up.

Digital twin

An AI that learns enough about you to act on your behalf with judgment.

AI agent

An AI system that takes goal-directed actions in the real world — not just answers in chat.

LLM (Large Language Model)

A neural network trained on text that produces text. The brain inside AI assistants.

Context window

The amount of text an LLM can process in one turn. Modern: 200K-2M tokens.

Voice AI

AI systems that handle spoken language — STT + LLM + TTS or unified models.

TTS (Text-to-Speech)

AI that turns text into natural-sounding voice. ElevenLabs leads in quality.

STT (Speech-to-Text)

AI that transcribes spoken audio into text. Deepgram + Whisper are the standards.

Tool use

An LLM calling external functions (APIs, code, data) instead of just generating text.

System prompt

Instructions given to an LLM at the start of every conversation that shape its behavior.

Embedding

A numerical vector representing the meaning of text. The basis of semantic search.

Semantic search

Search by meaning, not keywords. 'Insurance renewal' finds 'policy expiry'.

Structured output

Forcing LLMs to return JSON conforming to a schema, not free text.

BYOK (Bring Your Own Key)

You provide the LLM provider's API key; the app runs queries on your account.

MCP server (Model Context Protocol)

Standardized way for AI agents to discover and call external tools. Anthropic-led spec.

Composio

A managed service exposing 200+ apps as agent-callable tools with OAuth handled.

Prompt cache

Caching the static part of a prompt across turns to reduce cost and latency.

Circuit breaker pattern

When an external service fails, temporarily stop calling it instead of retrying. Used in Niyra's LLM failover.

OAuth token refresh

Refreshing access tokens before they expire — proactive vs reactive.

Vector memory

Storing facts as embeddings for semantic retrieval — Niyra's long-term memory layer.

Skill system

JIT-loaded prompt rules that surface per turn based on context — reduces hallucination.

Subagent

A specialized AI agent dispatched by a primary agent for a specific task. Enables parallelism.

Cron automation

Scheduled execution of agent tasks — natural-language schedules parsed to cron expressions.

Agentic loop

Plan → call tool → observe result → reflect → next step. The runtime pattern of every modern AI agent.

JIT loading (just-in-time)

Loading code or context only when needed, on the turn it's needed. Niyra's skill pattern.

Channel

Niyra on WhatsApp

Voice notes, photos, PDFs. Same Niyra as web. Where you already are.

Feature

She remembers

Vector memory across sessions. Tell her on Monday, she knows on Friday.

Solutions

AI for founders

The AI that runs the boring half of your company while you build.

Pricing

Free trial, no card

2,000 credits to put Niyra through a real day. Pro at $9 intro.

Integrations

32 apps connected

Gmail, Calendar, Slack, Linear, GitHub, Notion. All via OAuth.

Docs

5-minute quickstart

Sign up, connect Gmail, set a brief. Niyra running in 5 minutes.