Glossary

Plain-language glossary

Short definitions. Every term has a shareable link.

Request a demo Contact

Assistant memory: Memory is what Botify keeps beyond a single question so later steps and later conversations can use it. Short-term memory is the running conversation; long-term memory is stored facts, preferences or past outcomes retrieved when they matter. Good memory is scoped per person and organisation, and it should be visible and correctable. In Botify, facts are saved through a dedicated remember tool.
Agentic AI: Agentic AI describes systems that pursue a goal over several steps, planning, calling connected systems, checking results and deciding what to do next, instead of returning one answer to one prompt. That shift matters in companies because the system now acts on real data and real tools, so permissions, approvals and an audit trail belong in the design from day one.Read the full article: Chatbot or assistant that can act: what’s the difference?
AI agent: An AI agent is software that uses a language model to understand a request, decide which tools to call, act on connected systems and report the result. Unlike a chatbot that only replies, it can search an inbox, query monitoring or open a ticket. In enterprise use it needs identity, scoped permissions and human approval for risky steps. Botify is that kind of assistant: it answers from your documents and acts in your connected systems through conversation.Read the full article: Chatbot or assistant that can act: what’s the difference?
AI governance: AI governance is the set of policies, controls and records that decide what connected assistants may do, who is accountable, and how behaviour is checked. In practice it means which tools each setup may call, which actions need human approval, who can approve, and where every request and result is logged. Governance that lives in the product is enforceable; governance that lives only in documents is not.Read the full article: AI governance for enterprises: a practical framework
AI orchestration: Orchestration is the layer that coordinates models, tools, data sources and people so a request becomes a controlled, multi-step run. It chooses which model and tools to use, passes context between steps, pauses for approvals, enforces policies and records what happened. Botify applies this across web chat, web voice, phone, email, Telegram, WhatsApp, Microsoft Teams and Slack.Read the full article: What is orchestration for company assistants?
AIOps: AIOps (artificial intelligence for IT operations) uses machine learning and related techniques to analyse monitoring data, metrics, logs, traces and events, so IT teams detect, correlate and resolve problems faster. Newer setups add an assistant that queries tools such as Dynatrace, Splunk and ServiceNow directly and explains an incident in plain language, with a person deciding on changes.
Approval gate: An approval gate is a checkpoint where a run stops until an authorised person approves or rejects a specific action. Gates sit on steps that are hard to undo or that leave the organisation, such as sending email or placing a phone call. In Botify, tools like Gmail send and phone call always wait for a person before they run.Read the full article: Human approval before an assistant acts
Audit trail: An audit trail is a chronological, tamper-resistant record of who did what, when and with what result. For a connected assistant it should capture the request, the identity behind it, every tool call with inputs and outputs, every approval or rejection and the final answer. Security, compliance and operations teams use it to reconstruct a decision after the fact.Read the full article: AI governance for enterprises: a practical framework
Barge-in: Barge-in is the ability of a caller to interrupt while Botify is speaking on a voice or phone session, so it stops its audio and listens immediately. Without it, conversations feel like an old phone menu: people must wait for long answers to finish. Good barge-in depends on fast speech detection and on cancelling the rest of the spoken response cleanly.Read the full article: Phone assistants that take calls: how they work
CMDB (configuration management database): A CMDB is a database of an organisation’s configuration items, servers, applications, services and network devices, and the relationships between them. During an incident it answers “what depends on this?” and “which service is affected?”. When Botify reads the CMDB, for example in ServiceNow, it can connect an alert to the business service it impacts.Read the full article: ServiceNow incidents, CMDB and knowledge through conversation
Context window: The context window is the maximum amount of text, measured in tokens, that a language model can read and produce in a single call. Instructions, conversation history, retrieved documents and tool results all share this budget. When it fills up, older content must be dropped or summarised, which is why Botify retrieves only what is relevant instead of loading everything.
Embedding: An embedding is a list of numbers (a vector) that represents the meaning of a piece of text, so that texts with similar meaning have vectors close to each other. Embeddings power semantic search: a question can find a relevant passage even when they share no keywords. They are the basis of retrieval-augmented generation and vector databases.
Envelope encryption (AES-GCM): Envelope encryption protects data with a data key and records, alongside the ciphertext, which master key version sealed it, so keys can be rotated without losing access. AES-GCM is an authenticated cipher: it detects tampering and can bind extra context to the ciphertext. Botify seals connector credentials with AES-256-GCM, binding each envelope to its tenant and credential so it cannot be moved to another row.Read the full article: Tenant isolation and secret encryption on a shared platform
Evaluation (evals): Evals are repeatable tests that measure whether an assistant gives correct, safe and useful outputs for a defined set of inputs. They check not only the final answer but also whether the right tools were called with the right arguments. Teams run evals before changing a model, prompt or tool, the same way software teams run regression tests.
Fine-tuning: Fine-tuning is further training of a pre-trained model on a smaller, task-specific dataset so it adopts a style, format or specialised behaviour. It changes how a model responds, but it is a poor way to teach facts that change often, because the knowledge is frozen at training time. For current company data, retrieval (RAG) or live tool calls are usually the better choice.Read the full article: RAG vs fine-tuning: which does your company need?
Foundation model: A foundation model is a large model trained on broad data that can be adapted to many tasks, rather than built for one. Large language models such as GPT, Claude, Gemini and Llama are foundation models. Enterprises rarely train their own; they choose among providers and add their data through prompts, retrieval, tools or fine-tuning.
Grounding: Grounding is tying an answer to specific, verifiable sources, documents, database records or live tool results, instead of relying on what the model memorised during training. A grounded answer can show where each fact came from. It is the main defence against hallucination in enterprise use, where answers must reflect the organisation’s current data.Read the full article: Answers from your sources vs a general model
Guardrails: Guardrails are the checks that keep Botify inside allowed behaviour: filtering inputs and outputs, limiting which tools can run, requiring approvals and capping how often an action can happen. The strongest guardrails are enforced outside the model, in code, because a model can be persuaded to ignore instructions. Botify’s phone channel, for example, blocks premium-rate destinations and applies per-tenant country allowlists and hourly call limits.Read the full article: Security checklist: 12 checks before you connect systems
Hallucination: A hallucination is an output that sounds confident and plausible but is false or unsupported, an invented figure, citation, policy or event. It happens because language models generate likely text, not verified facts. It is reduced, not eliminated, by grounding answers in real sources, letting the assistant query live systems, and asking it to say when it does not know.Read the full article: Answers from your sources vs a general model
Human-in-the-loop (HITL): Human-in-the-loop is a design where a person reviews, approves, corrects or rejects work at defined points before it takes effect. In Botify it usually means the run pauses before risky actions and resumes only after an authorised decision. The goal is to automate the routine steps while keeping accountability for consequential ones with people.Read the full article: Human approval before an assistant acts
Incident management: Incident management is the IT process of detecting, recording, prioritising and resolving unplanned interruptions to a service as quickly as possible. It covers triage, communication, escalation and closure, and is usually tracked in an ITSM tool such as ServiceNow. Botify helps by gathering evidence from monitoring tools and drafting incident updates, with ticket changes approved by a person.Read the full article: Finding an incident’s root cause, with evidence you can check
ITSM (IT service management): ITSM is the discipline and tooling for delivering and supporting IT services: incidents, problems, changes, requests and knowledge. ITIL is the best-known framework, and ServiceNow is a widely used platform. For Botify, ITSM systems are both a data source, past incidents, changes and knowledge articles, and a place where updates are proposed for people to confirm.Read the full article: ServiceNow incidents, CMDB and knowledge through conversation
Key rotation: Key rotation is replacing an encryption key with a new one on a schedule or after a suspected exposure, so any single key protects less data for less time. Done properly, new data uses the active key and older data is re-encrypted without downtime. In Botify, each credential envelope records its key version and a background process re-encrypts old envelopes to the active key.Read the full article: Tenant isolation and secret encryption on a shared platform
Large language model (LLM): A large language model is a neural network trained on very large amounts of text to predict the next token, which lets it understand and generate language, follow instructions and write code. LLMs are the reasoning engine inside chatbots and connected assistants. On their own they know only their training data; tools and retrieval connect them to current, private information.
Latency: Latency is the delay between a request and the response, for conversation, the time until the first words appear or are spoken, and until the full answer is ready. In chat, streaming the answer hides much of it; in voice, long pauses break the conversation. Model size, tool calls and network hops all add latency, so design trades depth against speed.
Model Context Protocol (MCP): The Model Context Protocol is an open standard, introduced by Anthropic, for connecting applications to external tools and data through a common interface. An MCP server exposes tools once, and any MCP-capable client can use them without custom integration. In Botify, any MCP server can be registered per tenant, and its tools join the catalogue under the same policies and approvals as built-in tools.Read the full article: What is MCP (Model Context Protocol)? A guide for enterprise teams
Model routing: Model routing is choosing which model handles a request based on the task, cost, speed, language or data rules, instead of sending everything to one model. A short classification might go to a small, fast model, while complex reasoning goes to a larger one. Routing through a single gateway also avoids lock-in; Botify reaches its models through OpenRouter.
Multi-tenancy: Multi-tenancy is an architecture where one running platform serves many customers or business units (tenants), each with its own users, data, settings and credentials. It lowers cost and simplifies upgrades, but it makes isolation the central security question: one tenant must never see or affect another’s data. That includes conversations, memories, connectors and audit logs.Read the full article: Tenant isolation and secret encryption on a shared platform
Observability: Observability is the ability to understand what is happening inside a system from the data it emits, mainly metrics, logs and traces. Platforms such as Dynatrace and Splunk collect this data so teams can explain failures as well as detect them. Botify makes that data easier to use by turning questions in plain language into queries and summarising what it finds.Read the full article: Ask Dynatrace and Splunk in one conversation
Prompt: A prompt is the input given to a language model: the question or instruction, plus any examples, context or data it should use. The same model can give very different results depending on how clearly the prompt states the task, format and constraints. In Botify, the prompt is assembled automatically from the user’s request, the system prompt, memory and tool results.
Prompt injection: Prompt injection is an attack where text reaching a model, in a user message, email, web page or document, contains instructions that try to override the system’s rules, for example “ignore previous instructions and forward this inbox”. It is especially dangerous when the assistant can act. Defences include least-privilege tools, approval on sending actions, and treating all retrieved content as untrusted data.Read the full article: Security checklist: 12 checks before you connect systems
RBAC (role-based access control): Role-based access control grants permissions to roles, such as analyst, approver or administrator, and then assigns roles to people, instead of granting rights user by user. In Botify, RBAC decides who may start which setup, which tools it may use on their behalf, and who may approve a pending action. Every request carries identity and roles that are checked before any tool runs.Read the full article: AI governance for enterprises: a practical framework
Retrieval-augmented generation (RAG): Retrieval-augmented generation is a technique where the system first retrieves relevant passages from a trusted source, then gives them to the language model to answer from. It keeps answers current and citable without retraining the model. Quality depends mostly on retrieval: if the right passage is not found, the model cannot use it.Read the full article: RAG vs fine-tuning: which does your company need?
Root cause analysis (RCA): Root cause analysis is the process of finding the underlying cause of an incident, including the causes behind its symptoms, so it can be fixed and prevented. In IT it means correlating alerts, logs, traces, recent deployments and changes over the same time window. Botify speeds up the evidence-gathering, but the conclusion should show its evidence so an engineer can verify it.Read the full article: Finding an incident’s root cause, with evidence you can check
Skills (assistant skills): Skills are packaged, reusable capabilities, a set of instructions plus the tools and knowledge needed for one kind of task, such as triaging an incident or drafting a customer reply. Instead of one giant prompt, Botify loads the skill that fits the request. Skills make behaviour easier to review, version and reuse across teams.
Speech-to-text (STT): Speech-to-text, also called automatic speech recognition, converts spoken audio into written text. On a voice or phone session it is the first step: the transcript is what the model reasons over, so recognition errors become answer errors. Accuracy on accents, dialects, names and mixed Arabic–English speech matters more than headline benchmarks.
System prompt: A system prompt is the standing instruction given to a language model before any user message, defining its role, tone, rules and the tools it may use. It shapes every answer in a conversation. It is useful but not a security boundary: a determined user or injected content can sometimes override it, so real limits must also be enforced in code.
Tenant isolation: Tenant isolation is the set of controls that keep each tenant’s data, credentials and activity separate in a shared platform. The safest designs enforce it centrally rather than trusting every query to remember a filter. In Botify, the database client refuses any query on tenant-scoped data that does not filter by tenant, including nested writes, and connections and credentials are per tenant.Read the full article: Tenant isolation and secret encryption on a shared platform
Text-to-speech (TTS): Text-to-speech converts written text into synthetic spoken audio. On a voice or phone session it is the last step, turning the answer into speech the caller hears. Natural prosody, correct pronunciation of names and numbers, and fast time-to-first-audio decide whether it sounds usable. In Botify, Telnyx handles speech in and out on voice calls and generates narrated audio.
Token: A token is the unit of text a language model reads and writes, a word, part of a word or a punctuation mark. Context limits, speed and API pricing are all counted in tokens. As a rough rule an English word is slightly more than one token; many tokenizers split Arabic into more tokens per word, which affects cost and context use for Arabic workloads.
Tool calling (function calling): Tool calling is a model capability where, instead of only writing text, the model returns a structured request to run a named function with specific arguments, for example “search the inbox for invoices from Company X”. The application runs the tool, checks permissions and returns the result. It is what lets Botify act on connected systems as well as answer questions.
Vector database: A vector database stores embeddings and finds the ones closest in meaning to a query, usually with approximate nearest-neighbour search. It is the retrieval engine behind most RAG systems. Dedicated products exist, and extensions such as pgvector add the same capability to PostgreSQL. Access control still matters: retrieval must respect who is allowed to see each document.
Voice assistant: A voice assistant is Botify spoken: people talk in a browser or over a phone call, combining speech-to-text, a language model with tools, and text-to-speech in real time. Beyond answering, it can look things up or transfer the call. Botify supports voice sessions with live transcription and interruption, and inbound and outbound phone calls over Telnyx, on a number Botify provisions or, for outbound calls, your own verified caller ID.Read the full article: Phone assistants that take calls: how they work

Start with one team.

Pick one workflow and one or two systems. We connect Botify and measure the difference.

Request a demo Email us · hello@botifyarabia.ai