Plain-language definitions of the key terms, protocols and concepts behind Ethora’s chat, AI and app-building platform.
The A2A (agent-to-agent) protocol is an emerging open standard for how autonomous AI agents advertise their capabilities through an agent card, discover one another and exchange work. It complements MCP, which connects agents to tools, and is still early, so robust systems layer reputation and social memory on top of the self-published card.
Read full entry →A list attached to a resource that names who may access it and which operations they may perform. Covers file system, cloud, network and application ACLs, why precision trades off against scale, how ACLs combine with role-based access control, and how an admin panel uses them to grant team members granular access.
Read full entry →Video delivery that switches quality to match the viewer’s connection. The source is transcoded into an encoding ladder and cut into segments; the player picks the rendition it can sustain. Covers HLS and DASH, low-latency variants, and why WebRTC calls use simulcast instead of ABR.
Read full entry →The control layer that coordinates several AI agents, tools and steps into one workflow: planner-worker, router and supervisor patterns, handoffs, shared state, retries, turn limits and tracing. Names the common frameworks neutrally and shows how Ethora orchestrates agents inside chat rooms with a response gate, an autonomous-turn cap and a heartbeat scheduler.
Read full entry →What an AI agent remembers about other agents and people: their capabilities, trustworthiness, relationships and interaction history, so it can communicate, delegate and coordinate better over time. The missing piece that turns isolated agents into networks, and a space today’s single-user memory systems barely address.
Read full entry →How autonomous AI agents talk to each other to coordinate, delegate subtasks and combine results, the backbone of multi-agent systems. Unlike tool calling, agents are active parties, so reliable A2A needs identity, delivery guarantees, ordering and an audit trail, just like human chat.
Read full entry →Agentic AI describes AI systems that pursue goals autonomously by decomposing them into steps, calling tools, observing outcomes and self-correcting. Unlike a chatbot that responds to a single prompt, an agentic system runs multi-step loops and can write to external systems, making capability and responsible scoping both critical.
Read full entry →Technology that enables machines to mimic human cognitive functions such as learning, problem-solving and decision-making. In Ethora, AI refers to automated conversational agents powered by large language models — chatbots and embeddable AI widgets that let developers build intelligent, conversational interfaces without extensive coding.
Read full entry →AI agents go beyond chatbots: they perceive their environment, plan a sequence of steps, call external tools or APIs, and act autonomously to reach a goal. They are the building block of agentic workflows in customer support, operations, and research, and can be embedded in messaging platforms via frameworks like MCP.
Read full entry →An AI assistant helps individual users with questions, text drafting, summarisation and step-by-step guidance through a natural conversational interface. Modern assistants draw on language models and retrieval systems to give accurate, grounded answers, and can be embedded directly in your product’s chat or messaging layer.
Read full entry →A conversational program that understands free text and writes its own replies with a large language model, unlike a scripted bot. Covers how AI chatbots work (LLM, system prompt, retrieval, tools), how they differ from rule-based bots and AI agents, the main risks and controls, and how to build one on your own data.
Read full entry →The input and output controls wrapped around an LLM application: topic restriction, PII and PHI filtering, prompt injection and jailbreak defence, toxicity checks and limits on which tools an agent may call. Covers how guardrails are built, what they cannot do, and how Ethora applies them at the messaging layer through Trust and Safety.
Read full entry →A software development kit for adding language-model features to an application: agents, retrieval over your own data, tool calling and streaming. Covers the difference between client libraries that call model APIs and platform SDKs that supply the agent runtime, knowledge base and chat surface, and how Ethora’s AI SDK fits.
Read full entry →A system physically isolated from the internet and other networks, used in defence, critical infrastructure and some healthcare and finance environments. Covers what air-gapped deployment requires of chat and AI software: offline installs, internal certificates, local models instead of cloud LLM APIs, and alternatives to public push notification services.
Read full entry →An open standard protocol for enterprise message routing via a central broker. AMQP underpins reliable event-driven architectures with guaranteed delivery, flexible routing rules, and message persistence, making it a common backbone for backend service integration alongside real-time chat protocols like XMPP.
Read full entry →A set of rules and protocols that let different software applications communicate with each other. In Ethora, APIs let client applications interact with the backend server and integrate the platform with existing or custom-built systems, accessed via API keys from the Admin panel and documented through Swagger.
Read full entry →An API gateway is the front door to a backend system: a single entry point that validates tokens, enforces rate limits, routes requests to the right service and centralises logging. For chat and AI platforms it also handles WebSocket upgrades and long-lived connections, making protocol support a key selection criterion.
Read full entry →The process of connecting different software applications via APIs to enable seamless data exchange and functionality. In Ethora it is the route to deeper customization — syncing user accounts with legacy systems, or building a custom frontend on top of Ethora's backend infrastructure.
Read full entry →An API token is a secret string that clients include in API requests instead of a username and password. Tokens can be scoped to specific actions, rotated without changing account credentials, and revoked individually. In chat and AI platforms they come in two varieties: long-lived app tokens for server-to-server calls and short-lived user JWTs for client sessions.
Read full entry →In Ethora, an App is an entity with one or more frontend components connected to a single backend context. Each App is an isolated server-side container with its own users, files, chats and data, managed by its owner through the Admin Panel or built on with Ethora's SDKs.
Read full entry →The user or organization that signs up for an Ethora account and maintains a project on the platform. App Owners have full administrative control through the Admin Panel — managing apps, configuring settings, viewing statistics, and granting team members granular access via ACL settings.
Read full entry →An immutable, time-ordered record of who did what and when inside a system. Audit logs are a foundational requirement in HIPAA, SOC 2, ISO 27001 and PCI DSS: they give compliance teams evidence that access controls are enforced and let security teams reconstruct events after an incident. Most effective when stored outside the system they monitor, in write-once object storage.
Read full entry →The royalty-free video codec from the Alliance for Open Media, finalised in 2018. AV1 matches H.264 quality at roughly half the bitrate and beats VP9 and HEVC by about 30 percent, at the price of heavy encoding. Covers hardware support, scalable video coding and how AV1 is negotiated in WebRTC calls.
Read full entry →Everything that happens on the server side, out of sight of the end user. In Ethora the backend includes the API layer, chat server and application logic — a Backend-as-a-Service (BaaS) that lets teams focus on the frontend while leveraging pre-built server capabilities.
Read full entry →A backup policy defines backup frequency, retention schedule, storage location, and restoration testing. It is the foundation of any disaster-recovery plan. Cloud providers own and run the backup policy; self-hosted deployments give you full control over schedule, location, and retention, critical for HIPAA, financial record-retention, and other regulated-data requirements.
Read full entry →The data rate of an audio or video stream measured in bits per second. Bitrate is the primary lever for balancing quality against bandwidth. WebRTC calling systems adjust bitrate dynamically per participant, and simulcast allows senders to transmit multiple bitrate layers so each viewer receives the quality that matches their connection.
Read full entry →The vendor deploys and operates its software inside your own cloud account, so you own the infrastructure, region and data while the vendor handles upgrades and uptime. BYOC sits between SaaS and self-hosting. Ethora’s dedicated server works exactly this way: Ethora TechOps run it in your AWS, DigitalOcean or other account.
Read full entry →The contract HIPAA requires between a covered entity and any vendor that handles protected health information for it, including a chat platform that stores patient messages. Covers what a BAA must contain, who signs one, why it does not make you compliant on its own, and how it relates to a GDPR DPA.
Read full entry →A principle of distributed systems from Eric Brewer: during a network partition a system can guarantee either consistency or availability, not both. Because partitions are unavoidable, the practical choice is consistency versus availability, and systems are described as CP or AP.
Read full entry →A persistent, topic-named room in a chat workspace that members can browse and join. Channels differ from group chats in that they are discoverable and organised by subject rather than by a fixed invited list, making them the standard unit for team and community communication.
Read full entry →Two meanings, one name: the rounded container around each message in a conversation, and the floating launcher button that opens a website chat widget. Covers grouping, states, accessibility and launcher placement, and why both are exposed through theming. Ethora themes bubbles from config across web, React Native and Android.
Read full entry →The stored record of past messages in a conversation. Compares device-only and server-side history, explains multi-device sync and XMPP Message Archive Management, and shows how a server-side history with configurable retention and per-user erasure satisfies both record-keeping and privacy obligations.
Read full entry →A pre-built collection of libraries, APIs, UI components and documentation that lets developers add real-time messaging to their apps without building a chat system from scratch. It bundles native client libraries, server-side message routing and storage, ready-made UI, and real-time protocols — saving months of engineering.
Read full entry →An exported, readable record of a conversation for support, legal or compliance use. Covers what a trustworthy transcript must contain, the four common purposes, why exports should come from the server-side archive rather than a device, and how export actions themselves are logged.
Read full entry →The floating chat window on a website: a self-contained conversation UI, human or AI, embedded with a script tag or an iframe. Covers script versus iframe embeds, identity, proactive triggers and security review points. Ethora offers a one-line AI assistant widget, a WordPress plugin and a full chat widget SDK.
Read full entry →Software that holds a conversation with people by text or voice; in a chat app, a bot is an automated participant with its own profile. Explains what bot means in chat, rule-based versus AI versus hybrid bots, a short history from ELIZA to the GPT era, how bots work inside chat rooms and how to build one.
Read full entry →Concurrency is the number of users connected at the same moment. It is the key input to infrastructure sizing for chat and AI platforms, distinct from total registered users or monthly actives. Cloud plans cap it per tier; self-hosted deployments let you set and cluster to your own peak.
Read full entry →The policies, tools and workflows that keep user-generated content in a chat product safe and compliant. Content moderation covers automated classifiers, admin message deletion, user suspension and human review, and is a requirement for any regulated or public-facing chat environment.
Read full entry →The context window is the token budget an LLM works within for a single request. It covers the system prompt, conversation history, retrieved knowledge, and the model’s reply combined. Longer windows allow richer conversations and more RAG content but increase inference cost and latency.
Read full entry →Conversational AI lets software understand natural language and hold multi-turn dialogues rather than responding to rigid commands. It underpins customer support chatbots, in-app assistants, voice interfaces and AI agents, combining LLMs, intent recognition and context tracking to produce relevant, coherent replies.
Read full entry →An automated program that systematically browses and extracts content from websites by following links and sitemaps. It parses HTML while respecting server constraints and organizes the data into a searchable index — valuable for building knowledge bases that give AI systems current, business-specific information.
Read full entry →The percentage of customers who rate an interaction as satisfied or very satisfied, collected with a one-tap survey at the end of a chat. Covers the formula, benchmarks, response bias and how CSAT differs from NPS and CES. With Ethora you build the post-chat rating into the widget and can use AI sentiment as a proxy.
Read full entry →Claude Shannon’s 1948 model of communication and information theory: a message flows from a source through an encoder, across a channel that adds noise, to a decoder and destination, bounded by the channel’s capacity. It is the blueprint behind every chat, voice and AI messaging system.
Read full entry →The GDPR Article 28 contract between a data controller and the processor that handles personal data on its behalf. Covers the mandatory clauses, sub-processors, Standard Contractual Clauses, how a DPA compares with a HIPAA BAA, and what to check in a chat vendor’s DPA before your users’ messages depend on it.
Read full entry →The requirement that data be stored and processed only within a defined country or region. Data-residency rules in GDPR, HIPAA-adjacent frameworks, and national data-localisation laws drive many enterprises away from multi-tenant SaaS toward on-premise or dedicated server deployments where the physical location of every record is known and auditable.
Read full entry →The principle that data is governed by the laws of the place where it is stored and processed. Explains how sovereignty differs from residency and localisation, why GDPR transfer rules, Schrems II and the US CLOUD Act collide, what sovereign cloud offerings promise, and how a dedicated chat deployment in your own region answers the question.
Read full entry →DAU counts unique users who interact with a product on a single day. The DAU/MAU ratio reveals engagement depth. For infrastructure, DAU combined with session length converts to peak concurrent users, the real sizing input. Self-hosted teams calculate DAU from their own logs; cloud plans absorb the capacity planning automatically.
Read full entry →A machine or infrastructure-as-a-service instance where an application runs separately from multi-tenant environments, giving the owner greater control over maintenance, scalability and security. Organizations choose it for stronger isolation, operational control and better SLAs, at a higher cost than shared hosting.
Read full entry →The acknowledgement that a message reached the recipient’s device or server, and the middle rung of the sent, delivered, read ladder. Explains what each status means, how XMPP defines receipts and chat markers, why delivered is not the same as seen, and how receipts relate to typing indicators and presence.
Read full entry →A living virtual copy of a real asset, process, person or business. Covers the engineering origin (sensor-fed models of machines and buildings) and the AI meaning: a conversational agent grounded in an entity’s own documents with a defined persona, the pattern Ethora Agents uses to build twins on the AI SDK.
Read full entry →A technology or application that lets users store, access and manage digital assets. The term spans document wallets for personal or business records and crypto wallets based on cryptographic key pairs; in enterprise contexts, digital wallets often integrate messaging to streamline customer interactions.
Read full entry →A private conversation between exactly two users, visible to no one else. DMs are the core private-messaging layer in any chat product and are especially important in regulated contexts like healthcare and finance where conversations between two named individuals must stay confidential.
Read full entry →Disaster recovery defines how a service restores operations after a catastrophic failure. The two key metrics are RTO (how long recovery takes) and RPO (how much data can be lost). Cloud SLAs publish DR commitments; self-hosted deployments put you in control of the runbook, backup location, and drill schedule, a compliance requirement in healthcare and finance.
Read full entry →The processing that stops callers hearing their own voice bounced back from the other side’s loudspeaker. An adaptive filter learns the path from speaker to microphone and subtracts the predicted echo. Covers double-talk, delay estimation, line echo versus acoustic echo, and the canceller built into WebRTC.
Read full entry →A powerful open-source chat server from ProcessOne that implements the XMPP protocol for instant messaging. Built in Erlang, it provides reliable infrastructure for one-to-one and group chat at scale, though it typically needs custom extensions for user management, file attachments and push notifications.
Read full entry →Embeddings translate text, images or other data into numeric vectors where semantic similarity maps to mathematical closeness. They are the foundation of modern AI search, RAG pipelines and recommendation systems: content is embedded at index time and retrieved at query time by comparing vectors in a database rather than matching exact words.
Read full entry →Protecting data while it moves across the network with TLS 1.2 and 1.3, certificates, HSTS, WSS for WebSockets, TLS on XMPP streams and DTLS-SRTP for calls. Explains how transport encryption differs from encryption at rest and end-to-end encryption, and why regulated chat deployments usually pair strong transport encryption with server-side archiving rather than E2EE.
Read full entry →A method of securing messages so their content is readable only by the sender and recipient, with decryption keys held on the endpoints. It differs from encryption in transit (such as TLS) and encryption at rest, and it shapes which features can run on the server.
Read full entry →Messages that delete themselves after a timer or a view. Explains how disappearing messages are implemented, why they rarely vanish completely, why financial and healthcare regulators forbid them for work, and how policy-driven retention with targeted removal gives the privacy benefit without the record-keeping risk.
Read full entry →A functional programming language built for concurrent, distributed and fault-tolerant systems. Developed by Ericsson in 1986, it can support millions of concurrent processes, making it especially well-suited to messaging, telecom and real-time chat applications at scale.
Read full entry →The guarantee that every message is delivered and processed once, with no loss and no duplicates. Explains at-most-once, at-least-once and exactly-once semantics, why the network alone cannot provide the last one, and how chat gets there with acknowledgements, stream resumption (XEP-0198), stable stanza IDs (XEP-0359) and idempotent receivers.
Read full entry →A chatbot scoped to a fixed knowledge set, such as a help centre or policy documents, that answers repeat questions and hands off when it cannot. Compares keyword rules, intent classifiers and retrieval-augmented generation, covers deflection and fallback metrics, escalation to humans, common pitfalls, and how to build one on your own content.
Read full entry →A model where independently operated messaging servers interconnect over a shared protocol so users on one server can message users on another, the way email works across providers. XMPP and Matrix bring federation to real-time chat, and each operator keeps control of its own server and data.
Read full entry →Fine-tuning adapts a pre-trained language model to a specific domain or task by continuing training on a smaller labelled dataset. It sits between prompt engineering (no weight changes) and training from scratch, and is commonly used to specialise models for healthcare, legal, or customer support applications.
Read full entry →The client-side parts of an application that users see and interact with in a browser or on a device, as opposed to backend server-side operations. In Ethora, the frontend layer includes white-labelled iOS and Android apps, web apps and SDKs, so developers can deploy a UI without building it from scratch.
Read full entry →The llama.cpp model file format, successor to GGML, that packs quantised weights, tokenizer and metadata into one file for fast local inference. Explains the Q4_K_M, Q5_K_M and Q8_0 quantisation levels, where GGUF is used (Ollama, LM Studio, llama.cpp), how it differs from safetensors, and how a local GGUF model plugs into a chat platform.
Read full entry →Anchoring a language model’s answers in evidence it is shown at answer time, such as retrieved documents, tool results or live data, instead of training memory. Covers document, tool and structured grounding, citations and refusal, the relationship to RAG and hallucination, and how Ethora agents are grounded in a per-agent knowledge base.
Read full entry →Multi-party messaging where three or more participants share a room and all messages are visible to everyone in it. Group chat is the core pattern behind team channels, community spaces, care team coordination and any scenario where a shared conversation thread is needed.
Read full entry →gRPC is Google’s open-source RPC framework that sends binary Protocol Buffer messages over HTTP/2 instead of JSON over HTTP/1.1. The result is smaller payloads, lower latency and built-in streaming support. It is the dominant choice for service-to-service communication in microservice architectures, though REST and WebSocket remain better fits for browser and mobile client APIs.
Read full entry →The most widely deployed video codec, standardised in 2003 as ITU-T H.264 and MPEG-4 Part 10. Hardware encoders and decoders in nearly every device make it the default for battery-sensitive video calls, and RFC 7742 makes it mandatory in WebRTC browsers alongside VP8. Compared here with VP9, HEVC and AV1.
Read full entry →AI hallucination happens when a language model generates confident but factually wrong or invented content. It is one of the most important failure modes to mitigate in production, particularly in regulated industries. RAG grounding, source citation, and output validation are the main defences.
Read full entry →High availability keeps a service running through individual component failures using redundancy and automatic failover. Cloud plans provide HA by default within the shared infrastructure. Self-hosted deployments let you design and document your own HA architecture, a requirement for regulated-industry compliance audits.
Read full entry →Apple’s adaptive bitrate streaming protocol that powers large-scale live and on-demand video delivery. HLS breaks streams into short HTTP segments served from a CDN and auto-adapts quality to each viewer’s bandwidth. It is universally supported but adds seconds of latency, making WebRTC the right choice for interactive real-time calls.
Read full entry →Growing capacity by adding servers rather than a bigger server. Covers why stateless services scale out easily and stateful chat does not: persistent connections, presence, group fan-out, ejabberd clustering, MongoDB replica sets and sharding, SFU instances for calls, and why autoscaling connection-heavy services needs care.
Read full entry →Idempotency ensures that retrying the same API call or message send produces no additional side effects. It is implemented via client-generated idempotency keys that the server uses to detect and deduplicate repeated requests. In messaging, it prevents the classic “duplicate message” bug when a retry follows a network failure.
Read full entry →Real-time messaging built inside your own product rather than linked to a third-party app. In-app chat keeps users in context, supports text, media and AI agents, and lets the product owner control where the data lives. The preferred pattern wherever compliance or brand consistency matters.
Read full entry →AI inference is what happens every time a model generates a response: new input is passed through the model’s fixed weights in a forward pass to produce output. Inference cost, latency, and throughput are the primary engineering concerns when deploying language models at scale, especially in self-hosted setups.
Read full entry →Real-time text conversation over the internet with presence, from ICQ and AOL Instant Messenger to Jabber and the open XMPP standard. Covers how clients, servers and protocols fit together, how IM differs from email and SMS, and how in-app chat carries the same idea into products today.
Read full entry →The variation in packet arrival timing that causes choppy audio and video in real-time calls. Even with adequate bandwidth, high jitter disrupts smooth media playback. WebRTC engines handle jitter automatically with adaptive jitter buffers that smooth arrival variance while minimising added latency.
Read full entry →A compact, signed token that carries claims about a user, such as identity, roles and expiry. Made of a header, payload and signature, it lets a server verify a request without a session store, which is why it underpins stateless API, real-time and single sign-on authentication.
Read full entry →A knowledge base in AI is the curated document store that powers RAG: the agent searches it at inference time, retrieves relevant passages, and includes them in the model’s prompt. It reduces hallucination, keeps answers current without retraining, and can be scoped to private or regulated content.
Read full entry →Facts stored as entities and typed relationships, so software can follow connections and answer multi-hop questions. Covers triples, ontologies, graph databases, how a knowledge graph differs from a vector database, what GraphRAG adds, and how an Ethora agent reaches an existing graph through tool calling.
Read full entry →The delay between sending and receiving. Covers one-way delay versus round-trip time, mouth-to-ear and glass-to-glass latency in calls, where the milliseconds come from (distance, queueing, encoding, jitter buffers), the ITU-T G.114 target of 150 ms for voice, and the architecture choices that keep chat and calls fast.
Read full entry →The open protocol (RFC 4511) for reading and writing a directory of users, groups and credentials, most often Active Directory. Covers entries, DNs and attributes, bind and search, transport security, LDAP’s two enterprise roles (authentication and user sync), and how directory users reach Ethora through API provisioning.
Read full entry →A Large Language Model — an AI system trained on vast amounts of text with billions or trillions of parameters to understand and generate human-like language. LLMs excel at generation, translation, summarization, question answering and code, and their packed knowledge lets them act as capable assistants across many applications.
Read full entry →How to measure whether an LLM application is actually good: public benchmarks for choosing a model, task-specific golden sets for your own use case, LLM-as-judge, RAG metrics such as faithfulness and relevance, regression suites and human review. Includes a practical suite for a support chatbot and how to evaluate an Ethora agent through its one-shot reply endpoint.
Read full entry →A proxy between your applications and model providers that exposes one API and handles routing, fallbacks, credentials, rate limits, caching, cost tracking, logging and content filtering. Explains the OpenAI-compatible contract, the trade-offs, and how Ethora agents can point at a gateway or at a self-hosted model through the same endpoint setting.
Read full entry →Tracing and monitoring what an LLM application does in production: prompts, completions, tokens, cost, latency, tool calls and quality signals, linked in one trace per request. Covers what to record, the main tools, why prompt logs are personal data, and how Ethora’s transcript memory and audit trail keep a record of agent conversations.
Read full entry →A proposed convention (Answer.AI, 2024) for a Markdown file at a site’s root that tells language models what the site is and which pages matter. Covers the format, llms-full.txt, how it differs from robots.txt and sitemaps, what is and is not known about adoption, and why Ethora publishes one.
Read full entry →Load testing simulates high user volumes to measure where a chat or AI service slows down or breaks. Essential before any production launch, it feeds directly into SLO targets and capacity planning. Self-hosted teams run their own load tests; Ethora can help size and validate dedicated deployments.
Read full entry →A large language model that runs on your own device or server, keeping every prompt and response inside your network. Local LLMs power use cases where sharing data with a cloud API is not acceptable: healthcare assistants, financial document analysis, and other compliance-sensitive applications. Tools like Ollama and vLLM make deployment straightforward.
Read full entry →Internationalization (i18n) builds software so it can support any language and region; localization (l10n) adapts it to one. Covers the tooling, the difference between localizing a chat UI and translating the messages users send, and Ethora’s per-user chat language with real-time AI message translation.
Read full entry →How long logs are kept and why: regulatory minimums (HIPAA six years, PCI DSS twelve months, SEC 17a-4), GDPR storage limitation pulling the other way, and lifecycle rules that enforce the answer. Ethora offers configurable retention on its message archive, per-user erasure and immutable audit-log export to S3.
Read full entry →Long polling holds an HTTP request open at the server until there is something to report, then responds and the client immediately reconnects. It was the standard approach to real-time updates before WebSocket. Today it serves mainly as a fallback for environments that block persistent connections, and is implemented automatically by most real-time client libraries.
Read full entry →A parameter-efficient fine-tuning method that freezes a model’s weights and trains small low-rank adapter matrices, so customising a large language model fits on one GPU and ships as a file of megabytes. Covers QLoRA, swappable and merged adapters, cost, and when to choose LoRA over retrieval or prompting.
Read full entry →An open, federated messaging protocol from the Matrix.org Foundation in which every participating homeserver replicates a room’s history, with Olm and Megolm end-to-end encryption and bridges built in. Explains how it works, its adoption in European public-sector messengers, and a factual comparison with XMPP, the protocol Ethora runs on.
Read full entry →MAU counts unique users who interact with a product in a rolling 30-day period. Most managed chat platforms price tiers by MAU. Self-hosted deployments give you unlimited active users within a fixed infrastructure cost, removing the per-user billing variable.
Read full entry →A lightweight program that exposes a specific tool, data source or service to AI models via the Model Context Protocol. MCP servers let AI systems discover and use capabilities — tools for actions, resources for data, prompts for templates — without custom integration code, and typically run on your own infrastructure for security and vendor independence.
Read full entry →What Claude and ChatGPT call a remote MCP server once a user adds it under Settings, Connectors: a URL that speaks the Model Context Protocol plus the OAuth login or keyed URL that authorises it. Covers listed versus custom connectors, what a connector can access, how to remove one, and Ethora’s personal connector URL and OAuth endpoint.
Read full entry →User-controlled templates that a Model Context Protocol server publishes with a name, description and arguments, surfaced as slash commands or menu items and expanded with prompts/get. Can embed resources such as docs or code samples. Covers how prompts differ from tools, resources and system prompts, plus the six quickstart prompts the Ethora server ships.
Read full entry →The official open catalogue of Model Context Protocol servers at registry.modelcontextprotocol.io: metadata only, reverse-DNS names, a publishing CLI and a public API that other directories pull from. Covers what a listing contains, the vendor and community catalogues around it, and where the Ethora MCP server is published.
Read full entry →Read-only context that a Model Context Protocol server exposes by URI: files, documentation, schemas, API data. Application-controlled rather than model-controlled, fetched with resources/list and resources/read, with URI templates and optional change subscriptions. Covers how resources differ from tools and prompts and the four docs the Ethora server ships.
Read full entry →The action primitive of the Model Context Protocol. Each tool has a name, a JSON Schema for its arguments and annotations such as readOnlyHint and destructiveHint that let clients auto-approve reads and confirm risky calls. Covers tools/list, tools/call, listChanged and how many tools a client can handle.
Read full entry →The systematic, tamper-resistant retention of chat records for compliance, supervision, audit and eDiscovery. Covers the financial, healthcare and litigation rules that require it, the properties of a proper archive such as server-side capture and write-once storage, and why archiving and end-to-end encryption are alternatives rather than complements.
Read full entry →Middleware that routes messages between services using queues, topics, and routing rules. Brokers decouple producers from consumers, persist messages for reliability, and handle retries and dead-letter scenarios, forming the event backbone behind push notifications, webhooks, and AI job dispatch in chat platforms.
Read full entry →A temporary storage buffer that decouples message producers from consumers. Queues enable reliable async delivery with guaranteed processing, backpressure handling, and dead-letter management, and are widely used in chat infrastructure for push notification dispatch, webhook fan-out, and AI job scheduling.
Read full entry →An interface for sending and receiving messages from code, with two distinct meanings: channel APIs that reach SMS, WhatsApp, RCS and messenger platforms, and in-app chat APIs that provide users, rooms, real-time delivery and webhooks inside your own product. Explains the difference, how to choose, and where Ethora’s REST, XMPP over WebSocket, webhook and MCP interfaces fit.
Read full entry →The rules that define how a chat or messaging system formats, addresses, delivers and acknowledges messages. Examples include XMPP, MQTT, AMQP, Matrix and RCS; the choice shapes openness, scale, offline delivery, group semantics and security.
Read full entry →The carrier messaging standard that extends SMS with images, audio, video, and longer text. MMS routes media through carrier infrastructure and is widely supported in native messaging apps. RCS and in-app chat SDKs have largely superseded it for richer product and business messaging needs.
Read full entry →One repository for many packages: an SDK, its server, sample apps and tooling versioned and tested together. Covers workspaces, Turborepo, Nx, Changesets and git submodules, plus the trade-offs against one repo per package. The main Ethora repository is a monorepo holding the API, SDKs, bots and MCP server.
Read full entry →The ISO standard (ISO/IEC 23009-1) for adaptive streaming over HTTP. Covers the MPD manifest with its Periods, Adaptation Sets and Representations, CMAF segments shared with HLS, Common Encryption for DRM, low-latency DASH, and why iPhone still needs HLS while everything else can play DASH.
Read full entry →A lightweight publish and subscribe protocol built for constrained devices and unreliable networks. Clients publish to topics and a broker routes messages to subscribers, with three quality-of-service levels, making it the default choice for IoT and telemetry.
Read full entry →Several autonomous agents that perceive, act and interact with each other, cooperating or competing on tasks no single agent handles as well alone. Covers the classic distributed-AI meaning and the LLM-era meaning, what agents need to communicate reliably (identity, ordering, shared memory, audit), and how Ethora runs multi-agent rooms with per-room transcript memory.
Read full entry →An architecture where one application instance serves many customers simultaneously, with data kept logically separate between tenants. Multi-tenancy is efficient and cost-effective for most use cases. For regulated industries where logical isolation is not sufficient for auditors, a dedicated or on-premise deployment provides a physically separate data boundary.
Read full entry →Emoji responses attached directly to a chat message, showing at a glance how participants feel without a clutter of one-word replies. Reactions persist in real time across all participants and can be used as lightweight polls, status signals or triggers for backend automations.
Read full entry →Retrieval-Augmented Generation — an AI technique that makes large language models more accurate by grounding responses in your own data. Instead of relying only on training data, RAG retrieves relevant documents from a knowledge base and adds them to the model's context before it answers, so the system “looks up” current, specific information.
Read full entry →Rate limiting caps the number of API calls or messages a client can make in a given time window. When the limit is exceeded the server returns a 429 error. In chat platforms it protects against spam floods, runaway bots and abusive users. Self-hosted deployments can tune limits to match their specific traffic patterns.
Read full entry →The successor to SMS and MMS. RCS upgrades the phone’s native texting app with read receipts, typing indicators, high-resolution media, group chat, reactions and verified business messaging, and now works across iPhone and Android after Apple added support in iOS 18.
Read full entry →A popular open-source JavaScript library from Meta for building user interfaces, especially for the web. Its component-based architecture and virtual DOM enable efficient rendering and reusable, stateful UI elements, and it extends to native iOS and Android apps through React Native.
Read full entry →Functions with a use prefix, such as useState and useEffect, that give React function components state, side effects and reusable logic. Covers the core hooks, React 19 additions, the rules of hooks, custom hooks, and how a hooks-based app plugs into a real-time chat component that manages its own connection.
A read receipt tells a message sender that the recipient has opened their message. It is separate from a delivery receipt, which only confirms arrival. In regulated products, read receipts can also serve as compliance evidence that a critical message was seen.
Read full entry →A Model Context Protocol server that runs at a URL over Streamable HTTP instead of locally over stdio, so web assistants and hosted agents can use it with nothing installed. Covers the identity model (OAuth 2.1, PKCE, RFC 9728, API keys), session ids, discovery, abuse controls, hosting trade-offs, and Ethora’s hosted and self-hosted endpoints.
Read full entry →An authorisation model where permissions are attached to roles and users inherit them by holding a role. Explains users, roles, permissions and sessions, role hierarchies, how RBAC differs from ACLs and attribute-based control, and how roles apply in an admin panel and inside chat rooms.
Read full entry →Adobe’s low-latency ingest protocol that encoders use to push live video to media servers. RTMP remains the standard upstream format for platforms like YouTube Live and Twitch. Viewers receive HLS or WebRTC streams rather than RTMP directly, as browser playback via Flash ended in 2020.
Read full entry →A network control protocol for streaming media sessions. RTSP issues VCR-style commands (Play, Pause, Teardown) to a server while the actual video travels over RTP. It is the dominant protocol for IP cameras and CCTV systems and is typically transcoded to HLS or WebRTC for browser delivery.
Read full entry →A focused reply chain attached to a single message in a channel or room. Threads keep sub-discussions organised and prevent a busy main feed from becoming unreadable, making them essential in team tools, support queues and any high-volume community chat.
Read full entry →Technical support structures issue resolution into expertise tiers (L1 to L3) and severity levels (P1 to P4), with SLA response-time commitments for each. Cloud plans bundle support tiers by price; dedicated Ethora deployments include named contacts and L3 engineers who know your specific environment, often a procurement requirement for regulated-industry buyers.
Read full entry →Defining an app’s colours, fonts, spacing and light or dark mode as design tokens that components read, so one config restyles the whole interface. Covers tokens, per-platform implementation, dark mode, the roles that matter in a chat UI, and how Ethora applies deep theming from config across web, React Native and Android.
Read full entry →Throughput is the number of messages or requests a system processes per second. It determines how many simultaneous users a chat or AI platform can serve without degradation. Cloud plans publish throughput tiers; self-hosted teams size their own infrastructure with load testing.
Read full entry →Two meanings that matter to chat builders. In language models, tokenization splits text into the tokens that set context limits, pricing and speed (BPE, WordPiece, SentencePiece). In security, it replaces sensitive values such as card numbers or health identifiers with harmless substitute tokens. One line on blockchain asset tokenization for disambiguation.
Read full entry →Tool calling (also called function calling) lets an LLM invoke external functions, APIs, or services mid-conversation. The model emits a structured request, the application executes it, and the result feeds back into the response. It is the core mechanism that turns a chat model into an agent capable of taking real-world actions.
Read full entry →Decoding media and encoding it again with a different codec, resolution or bitrate. Explains why an MCU transcodes every call and an SFU never does, where transcoding still belongs in recording and HLS or DASH streaming pipelines, and what it costs in CPU and GPU capacity.
Read full entry →The function, and in a chat platform the server-side system, that protects users and the business from harm that happens through messaging: abusive language, unsafe content, fraud and leaks of personal or regulated data. The Ethora Trust & Safety system scans every message in real time at the chat server level and blocks, censors, quarantines or reports it by severity, using stop words, regular expressions and AI auto-moderation.
Read full entry →A typing indicator shows that someone in a conversation is currently composing a message. It reduces reply-uncertainty, improves perceived responsiveness, and originates from the XMPP XEP-0085 Chat State Notifications standard. Ethora’s Chat SDK handles the full composing, paused and gone lifecycle automatically.
Read full entry →How real-time media gets through routers and firewalls. STUN discovers a device’s public address, TURN relays when no direct path exists, and ICE tests every candidate pair to pick one that works. Explains cone versus symmetric NAT, UDP-blocking firewalls, and why an SFU with a public address makes traversal a one-sided problem.
Read full entry →The field of computer science that lets machines read, understand and generate human language. Covers the classic tasks (tokenization, entity recognition, sentiment, intent, translation, summarisation), the path from rules to statistics to transformers and LLMs, the difference between NLP, NLU and NLG, and where NLP shows up in chat products.
Read full entry →Audio processing that strips background sound from the microphone signal before it is sent, so the far end hears speech rather than fans, traffic and keyboards. Covers spectral subtraction and Wiener filtering, AI models such as RNNoise, the WebRTC suppressor browsers enable by default, and when to switch it off.
Read full entry →A reusable software package published to the Node Package Manager that encapsulates functionality or UI for JavaScript/TypeScript projects. These packages work across frontend frameworks and Node.js backends, shipping pre-built functionality with configuration, documentation, versioning and TypeScript support so developers don't rebuild features from scratch.
Read full entry →The identity layer on top of OAuth 2.0 that lets an application verify who a user is through an identity provider and receive a signed ID token. Covers the ID token and its claims, the authorization code flow with PKCE, discovery, how OIDC compares with SAML, and how Ethora’s social sign-on and enterprise SSO use it.
Read full entry →A communication approach where every channel a customer uses, including in-app chat, email, SMS and social messaging, is connected so context and history follow them from one to the next. Omnichannel differs from multichannel (simply being present on many channels) by requiring integration at the conversation layer. A Chat SDK typically anchors the in-app channel that ties the rest together.
Read full entry →Software that runs on servers owned and operated by the customer rather than in a vendor’s shared cloud. On-premise deployment gives regulated organisations direct control over where their data lives, how the network is configured, and who can access the system. For chat and AI workloads, it is the gold standard for data isolation in healthcare, finance and government.
Read full entry →The share of packets that never reach their destination. Packet loss comes from congestion, weak Wi-Fi and faulty hardware; chat protocols retransmit and stall, while WebRTC calls conceal the gaps with NACK, forward error correction and Opus loss concealment. Covers causes, measurement and what counts as an acceptable rate.
Read full entry →Devices exchanging data directly instead of through a server. WebRTC starts every one-to-one call this way, but NAT, group size and compliance push production calling onto a server such as an SFU. Explains mesh versus SFU, when a TURN relay quietly ends the P2P model, and why archiving needs a server.
Read full entry →Messages fixed to the top of a room or channel so important information survives the scroll. Covers what pinning does and does not do, typical uses, who is usually allowed to pin, how pins compare with threads and announcement channels, and what a pin is from an engineering point of view.
Read full entry →The sanctioned channel between a page and an iframe from another origin. Covers the syntax, the message event, explicit target origins, origin allow-lists and the two mistakes behind most postMessage vulnerabilities. Ethora’s iframe-based assistant widget is the kind of embed this API exists for.
Read full entry →Presence tells participants in a messaging system whether a contact is online, away, busy or offline right now. It is one of the foundational real-time features in any chat product and is especially valuable in operational contexts like healthcare or customer support, where knowing who is reachable before sending a message matters.
Read full entry →Cloud infrastructure dedicated to a single organisation, whether in its own data centre or hosted for it, combining cloud automation with exclusive hardware. Covers the NIST deployment models, public versus private versus hybrid, why regulated teams choose it, and how Ethora’s self-hosted and dedicated options fit inside one.
Read full entry →A large language model running entirely on your organisation’s own infrastructure, whether on-premises hardware or a private cloud tenancy. A private LLM ensures that sensitive data, including patient records, financial documents and proprietary business context, never passes through a third-party model provider. Often paired with RAG and a dedicated server for a fully self-contained AI stack.
Read full entry →The chat starts the conversation: a widget opens or an agent sends the first message when a trigger fires, such as time on the pricing page, exit intent or a scheduled check-in. Covers trigger types, etiquette and measurement. Ethora AI agents can speak first in rooms through a heartbeat scheduler.
Read full entry →A web application that installs to the home screen or desktop, works offline and receives push notifications, built from a web app manifest, a service worker and HTTPS. Covers what PWAs can and cannot do compared with native apps, iOS web push, and shipping an Ethora web chat as an installable app.
Read full entry →Prompt injection attacks embed malicious instructions in user messages or retrieved content to override an LLM’s system prompt. It is a key security concern for AI agents that process untrusted input, and it requires defence-in-depth: input filtering, minimal tool permissions, and action confirmation steps.
Read full entry →Any individually identifiable health information held by a HIPAA covered entity or business associate, in any medium. Covers the identifier groups behind the 18 Safe Harbor identifiers, why chat messages and attachments count, and the safeguards a messaging platform needs before it can carry PHI.
Read full entry →A messaging pattern where publishers send to named topics and subscribers receive through a broker, so the two sides stay decoupled. Its one-to-many, push-based model powers real-time chat fan-out, live feeds, event-driven microservices and IoT telemetry.
Read full entry →Messages delivered to a device even when the app is closed, sent through platform services such as Apple Push Notification service and Firebase Cloud Messaging. Each device has a unique token your backend addresses, and notifications can be visible alerts or silent, data-only syncs.
Read full entry →How networks and applications keep real-time traffic usable. Network QoS marks packets with DSCP so switches prioritise voice and video; application QoS adapts bitrate, buffers jitter and repairs loss. Covers RFC 2474, the WebRTC markings in RFC 8837, MOS scores and why chat degrades more gently than calls.
Read full entry →Quantization compresses AI model weights to lower numerical precision, shrinking memory requirements and speeding up inference. A model that needs 14 GB of GPU memory at 16-bit can often fit into 4 GB at 4-bit, making self-hosted LLMs practical on standard server hardware.
Read full entry →The XML-based standard behind enterprise single sign-on. SAML lets an identity provider, such as Okta or Azure AD, assert a user’s authenticated identity to connected applications, so employees log in once and gain access to everything. SAML integration is a common enterprise procurement requirement, especially in regulated industries where centralised user lifecycle management is a compliance control.
Read full entry →The IETF standard (RFC 7643 and 7644) for automatically creating, updating and deactivating user and group accounts in applications from an identity provider. Covers the REST and JSON protocol, why deprovisioning is an audit control, the JIT and custom-API alternatives, and Ethora’s API-provisioned end-user model.
Read full entry →Capturing a screen, window or browser tab as a live video track in a call. Covers the getDisplayMedia API and its picker, content hints for text versus motion, why mobile browsers cannot capture but native apps can, and the privacy habits that matter when the screen holds patient or account data.
Read full entry →A packaged set of libraries, APIs, tools, sample code and documentation for building on a platform. How an SDK differs from an API, a library and a framework, what to look for in a messaging SDK, and the Chat SDK, AI SDK, npm packages and MCP server that Ethora ships.
Read full entry →Chat that regulated organisations can trust: encrypted in transit and at rest, gated by authentication and access controls, recorded in an audit trail and retained or erased by policy. Explains the layers behind the label and the choice between end-to-end encryption and the compliance-archive model.
Read full entry →Semantic search retrieves content by meaning, not by matching exact words. It uses AI embeddings to represent queries and documents as vectors, then finds the closest matches in a vector database. The result is search that understands synonyms, paraphrases and intent, and it is the retrieval layer that makes RAG pipelines accurate and AI assistants genuinely useful.
Read full entry →MAJOR.MINOR.PATCH with rules about which number moves: SemVer 2.0.0 lets a version number say whether an update breaks, adds or fixes. Covers caret and tilde ranges in npm, pre-release tags and how calendar versioning compares. Ethora’s packages use year.month.patch numbers that stay valid SemVer.
Read full entry →Automatically reading the emotional tone of text as positive, negative or neutral, with intensity and specific emotions in richer forms. Covers lexicon, classifier and LLM approaches, how support chat uses it for routing and CSAT, its limits, and how an Ethora agent scores tone while Trust & Safety enforces policy.
Read full entry →Server-Sent Events (SSE) lets a server stream a continuous feed of text events to a browser over plain HTTP. The browser’s built-in EventSource API handles reconnection automatically. SSE is unidirectional (server to client only), which makes it ideal for AI token streaming, activity feeds and live notifications, while WebSocket is the better choice for bidirectional chat.
Read full entry →Generating a page’s HTML on the server for each request, then hydrating it in the browser. Covers SSR versus client-side rendering, static generation and streaming, the hydration mismatch rule, and why real-time chat components such as Ethora’s React component are rendered client-only inside a Next.js app.
Read full entry →A background script the browser runs apart from any page: it caches for offline use, intercepts requests and receives push messages when no tab is open. Covers the lifecycle, scope, HTTPS requirement and web push. Ethora delivers web push through Firebase Cloud Messaging via a service worker, with self-serve credentials in the admin panel.
Read full entry →The media server architecture behind scalable group video calling. An SFU receives one stream from each participant and routes individual streams to each other participant without re-encoding, keeping server CPU low. Per-subscriber simulcast layer selection means every participant gets the quality their network can support.
Read full entry →A private DOM subtree with scoped CSS attached to an element, so a component’s markup and styles stay separate from the page. Covers shadow roots, slots, open and closed modes and why it isolates styles but not scripts. Ethora’s assistant widget uses an iframe for the stronger boundary that sensitive conversations need.
Read full entry →Horizontal partitioning: data or traffic is split by a shard key across servers so capacity grows by adding machines. Covers shard-key design, MongoDB sharded clusters and why chat systems partition by room and by app. Ethora’s App containers and room model give a natural key on its ejabberd, Node.js and MongoDB backend.
Read full entry →Security Information and Event Management: the system that collects, normalises, correlates and retains logs from across an organisation to detect threats and satisfy audit requirements. Covers how a SIEM works, why regulation makes it mandatory, and how Ethora’s audit trail export and Trust & Safety reports feed one.
Read full entry →A WebRTC adaptive quality technique where a publisher transmits the same video at multiple resolutions and bitrates simultaneously. An SFU selects the right layer for each viewer based on their network, enabling every participant in a group call to receive the best quality their connection can support.
Read full entry →One customer, one dedicated instance: servers, database and configuration shared with nobody else. Single-tenant deployments give regulated buyers a physical data boundary, independent configuration and a small blast radius, at a higher operating cost than multi-tenant SaaS. Ethora’s dedicated server is single-tenant; its managed cloud is multi-tenant with isolated Apps.
Read full entry →The signalling protocol behind VoIP calls and video conferencing over IP. SIP handles session setup, codec negotiation, and teardown while media travels over RTP. It underpins enterprise PBX and carrier VoIP, and bridges to WebRTC-based calling through gateways.
Read full entry →A Service Level Agreement — a formal contract between a service provider and customer that defines the expected level of service with measurable standards and consequences for falling short. SLAs set clear expectations for availability and performance (such as uptime and support response times) and specify remedies like service credits when targets are missed.
Read full entry →An SLO is an internal reliability target that sits behind an SLA, covering metrics like uptime percentage, latency percentiles, and error rates. Cloud providers own the SLOs backing their SLAs; self-hosted deployments let you define and own SLOs that match your regulatory reporting requirements.
Read full entry →A compact language model, typically under about ten billion parameters, that runs on a single GPU, a laptop or a phone. Covers examples (Phi, Gemma, small Llama, Mistral 7B), quantization, local serving with Ollama and vLLM, and why regulated teams pair an SLM with retrieval on infrastructure they own.
Read full entry →The cellular carrier protocol for sending text messages without a data connection or app. SMS has near-universal reach and is widely used for OTP codes and transactional alerts, but lacks encryption, read receipts, and rich media. In-app chat provides a richer, more controllable alternative for ongoing product communication.
Read full entry →An independent CPA firm’s attestation report on a service provider’s controls against the AICPA Trust Services Criteria: security, plus optional availability, processing integrity, confidentiality and privacy. Explains Type I versus Type II, how SOC 2 relates to ISO 27001 and HIPAA, and how to read a vendor’s report during due diligence.
Read full entry →The encrypted profile of RTP defined in RFC 3711. SRTP adds AES encryption, an authentication tag and replay protection to every voice and video packet. WebRTC keys it with DTLS-SRTP and forbids plain RTP; through an SFU it protects each hop rather than the whole path.
Read full entry →Social Sign-On — an authentication method that lets users log in with existing credentials from services like Google, Facebook, Apple or LinkedIn. Built on OAuth 2.0, it authenticates through trusted identity providers, removing the need for separate usernames and passwords while improving both convenience and security.
Read full entry →How an application stores, updates and shares its data across components. Compares context, Zustand, Redux Toolkit and server-state caches, explains why real-time chat with messages, unread counts and presence needs a store discipline, and where an SDK that manages its own connection state fits alongside your app’s store.
Read full entry →The Model Context Protocol transport for network servers since revision 2025-03-26: one endpoint, JSON-RPC over POST with JSON or Server-Sent Events replies, an optional GET stream, the Mcp-Session-Id header and resumable streams. Explains why it replaced the two-endpoint HTTP+SSE transport and how the hosted Ethora server uses it.
Read full entry →The NAT traversal protocols that make WebRTC calls reliable across firewalls and NAT routers. STUN discovers the public IP and port each peer presents to the internet; TURN acts as a media relay when direct connections fail. Together they ensure video and voice calls connect even on strict corporate or mobile networks.
Read full entry →The standing instruction a language model reads before any user message: its role, tone, rules, scope and tools. Covers what belongs in a system prompt, why it is not a security boundary (prompt injection), practical prompt craft, and per-agent prompts in Ethora’s AI SDK.
Read full entry →A set of prebuilt, reusable interface components that share one design. For chat that means the room list, message list, composer, threads and media views. Covers what a complete chat UI kit includes, theming and escape hatches, and the React and React Native kits Ethora ships on its open-source backend.
Read full entry →The badge number on a channel or conversation showing how many messages a user has not yet read. Unread counts are the primary re-engagement signal in chat products and need server-side computation to stay consistent across multiple devices. They feed push notification badges and drive users back to important conversations.
Read full entry →Uptime measures what percentage of the time a service is available. The nines range from 99% (two nines, ~88 hours of allowed downtime per year) to 99.999% (five nines, ~5 minutes). Cloud SLAs publish the target; self-hosted deployments let you own and design for your own uptime commitment.
Read full entry →Vector databases store content as numeric embeddings and retrieve the most semantically similar entries to a query in milliseconds, even across millions of records. They are the retrieval backbone of RAG pipelines: documents are embedded at index time, and the closest chunks are fetched at query time to ground LLM responses in accurate, up-to-date information.
Read full entry →The compression algorithm that makes video transmission practical. Codecs like H.264, VP8, VP9, and AV1 encode at the sender and decode at the receiver, each offering different trade-offs of quality, bitrate efficiency, encoding CPU cost, and hardware support. WebRTC negotiates the best mutually supported codec automatically.
Read full entry →A programmable service for adding multi-party voice and video calls to your own app: rooms, access tokens, media tracks, an SFU and client SDKs on top of WebRTC. Covers what to expect from one, build versus buy, and the BAA, encryption and access requirements for HIPAA telehealth.
Read full entry →A logically isolated private network inside a public cloud such as AWS, Google Cloud or Azure, with the customer’s own address ranges, subnets, firewall rules and connectivity. Covers public and private subnets, gateways and private endpoints, what a VPC does and does not guarantee, and why Ethora’s dedicated deployments run inside the customer’s VPC.
Read full entry →An open-source serving engine that makes running large language models on your own hardware practical at production throughput. vLLM’s PagedAttention technique and OpenAI-compatible API let teams swap a cloud model endpoint for a self-hosted one with minimal code changes, a key building block for private LLM deployments in regulated industries.
Read full entry →Live spoken conversation inside an app, from gaming party channels to audio-only clinical check-ins. Explains the room model versus a dialled call, push-to-talk versus open mic, the Opus codec and the echo cancellation and noise suppression that run before encoding, and how Ethora offers audio-only calls on WebRTC.
Read full entry →Voice over IP carries phone calls as packets over the internet instead of over a reserved circuit on the telephone network. Covers signaling with SIP, media over RTP, codecs such as Opus, how WebRTC brought VoIP into browsers and apps, and why in-app calling needs no phone numbers at all.
Read full entry →A technology application that lets users securely store, organize and manage digital assets. A document wallet is a secure container for IDs, contracts and certificates that a user can store, share and digitally sign; unlike generic cloud storage it is purpose-built for verifiable documents and can use blockchain for immutable audit trails and cryptographic signatures.
Read full entry →A webhook delivers real-time event notifications to your server by sending an HTTP POST request the moment something happens in an external system. You register a URL, and the platform calls it with a JSON payload describing the event. Webhooks eliminate polling and are the standard integration pattern for connecting chat platforms to CRMs, analytics and AI pipelines.
Read full entry →The open standard and browser API for real-time audio, video and data shared directly between browsers and apps without plugins. Built from getUserMedia, RTCPeerConnection and data channels, with STUN/TURN and SFUs for connectivity and group calls, it powers most in-browser voice and video.
Read full entry →The setup exchange WebRTC leaves to the application: SDP offers and answers, DTLS fingerprints and ICE candidates carried over a channel the app chooses. Covers JSEP and trickle ICE, WebSocket versus XMPP Jingle versus SIP over WebSocket, what else signaling carries during a call, and why a chat server is the natural signaling layer.
Read full entry →A protocol (RFC 6455) that keeps a single connection open so a client and server can exchange messages in both directions at any time. Its full-duplex, low-latency model powers chat, live notifications, presence and collaborative apps, unlike request-and-response HTTP or one-way Server-Sent Events.
Read full entry →A customizable software application that a business can brand and tailor as its own without building it from scratch. The name comes from the idea of a blank label a company fills with its own branding before reselling or integrating the product — letting businesses launch fully branded apps in minutes rather than months.
Read full entry →Write Once, Read Many: storage that accepts a record and then blocks every change or deletion until its retention period ends. S3 Object Lock, compliance and governance modes, legal holds, and the rules (SEC 17a-4, HIPAA integrity) behind them. Ethora exports its audit trail as immutable logs to S3 for exactly this purpose.
Read full entry →An open, decentralized and extensible messaging protocol for real-time communication and presence. Originally known as Jabber and evolving since 1999, it is widely used in chat apps, online gaming and IoT, and can also act as a signaling channel for features like video calls and screen sharing.
Read full entry →A model performing a task it was given no examples of, from a description alone. Explains the zero-shot, few-shot and fine-tuning spectrum, why large language models are strong zero-shot learners, prompt patterns that raise accuracy, and the cases (private knowledge, subtle labels, strict formats) where zero-shot breaks down.
Read full entry →No terms match your search. Try another word.
From white-labelled apps to self-hosted chat servers and AI agents — talk to our team about the right setup for your project.
