Communication

Socket.IO63.2k★

Socket.IO enables real-time, bidirectional communication between browsers and servers for chat, live dashboards, multiplayer games, and collaborative editing. It wraps WebSockets (the protocol that keeps a persistent connection open between client and server) and adds automatic reconnection, room-based broadcasting, and fallback transports when WebSockets aren't available. The entire thing is free and MIT licensed. No cloud tier, no premium features, no gotchas. You install it, you use it. The docs are solid and the ecosystem is massive; every language has a compatible server implementation. Self-hosting is trivial if you're already running a Node.js server. It's literally a package you add. If you're scaling to thousands of concurrent connections, you'll need a Redis adapter to sync across multiple server instances, but that's a well-documented pattern. Solo developers: use it. Small teams: use it. Large teams: use it, but look at your architecture; at serious scale, you might want a dedicated message broker like NATS for backend-to-backend communication and Socket.IO just for the client layer. The catch: Socket.IO adds overhead compared to raw WebSockets. If you're building something latency-sensitive like a trading platform, go raw. For 95% of real-time web apps, it doesn't matter.

Discourse47.9k★

Discourse is forum software for the long conversations that chat buries: topics, replies, search, and categories, plus built-in chat, themes, and plugins. It has run communities for over a decade. GPL v2 and free to self-host, including the AI and Data Explorer plugins that ship in the repo. Self-hosting is one Docker container that bundles Postgres and Redis, on a Linux server with 1 GB of RAM minimum and 2 GB recommended. Docker is the only supported install, so no cPanel and no shared hosting. Upgrades are a button in the admin panel, but releases land monthly. Pin the extended support release if nobody wants that chore. You also need an SMTP provider for digests and email replies. Hosted pricing: a free plan with 2 staff seats and 10 categories, Pro at $100/mo, Business at $500/mo, Enterprise custom. Custom domains start at Pro. OIDC single sign-on starts at Business. SAML and custom plugins are Enterprise only. Solo and small communities: self-host or take the free hosted plan. Medium: Pro. Large: self-host with a real owner, or Enterprise. The catch: hosted plans below Enterprise cannot install their own plugins, so self-hosting is the only cheap route to customization. And Discourse says it plainly: hosting Rails applications is complicated.

Rocket.Chat46.1k★

Rocket.Chat is a self-hosted messaging platform where you control the data and no vendor decides to change pricing on you. battle-tested, used by government agencies and enterprises that can't put conversations on someone else's cloud. You get channels, DMs, threads, video calls, file sharing, and a marketplace of integrations. It works. It looks and feels like Slack circa 2020, not as polished, but functional. Self-hosting is free with the Community Edition. The Enterprise plan starts at $7/user/mo (billed annually) and adds features like read receipts, message auditing, LDAP/SAML sync, and premium support. Their cloud-hosted option starts at $7/user/mo as well. The catch: self-hosting Rocket.Chat is not trivial. It needs MongoDB, a Node.js server, and real maintenance. Updates can break things. The Community edition is missing features that teams expect from Slack, like proper SSO integration (SAML/LDAP is Enterprise only). And frankly, if your team is under 15 people, Slack's free tier might be less work.

Novu40.0k★

Novu gives you one API for every notification channel: email, SMS, push, in-app, and chat, so you stop duct-taping Twilio, SendGrid, Firebase, and Slack together yourself. You define a notification workflow once and it routes to the right channel based on each subscriber's preferences. The self-hosted open-source version is free and includes the core workflow engine, template management, subscriber preferences, and all the channel integrations. It runs via Docker Compose with MongoDB, Redis, and the Novu workers, an hour or two to set up and a moderate ongoing ops burden since you're running a multi-service stack with queues and a database. Cloud is metered by workflow runs now, not events. Solo developers: the free cloud tier (10,000 runs/month) covers most side projects. Small teams: free tier still works at low volume, or Pro from $30/mo. Growing teams: self-host to dodge per-run costs, or pay $250/mo for the Team tier to skip the ops. The newer versions also lean into an AI agent runtime, so Novu is drifting from pure notifications toward agent messaging. The catch: Novu has rearchitected more than once. It's now on a v3, code-first line that's cleaner but has broken backward compatibility along the way. If you built on older patterns, expect migration work, and keep an eye on how far the AI-agent pivot pulls the core product.

Mattermost39.1k★

Channels, threads, file sharing, integrations, video calls. Mattermost looks like Slack because Slack is what it exists to replace, and the free self-hosted edition covers messaging, file sharing, webhooks, slash commands, and basic access controls for as many users as you care to add. Self-hosting is Docker Compose or Kubernetes plus Postgres or MySQL. A small VPS carries a few hundred users without complaint. Releases land monthly, so budget 3 to 5 hours a month for upgrades, database backups, and the occasional plugin that breaks. Ops burden: moderate. The paid tiers restructured this year. Professional now includes both AD/LDAP sync and SAML SSO (single sign-on, the thing that lets your whole company log in with one corporate account), which used to mean buying Enterprise. Enterprise adds compliance export, eDiscovery, and legal hold. A new Enterprise Advanced tier sits above that for attribute based access control and mobile biometrics. Mattermost also stopped publishing per seat prices: licenses are prepaid annual seats sold through sales, and outside trackers still put Professional around $10 per user per month. Solo and small teams: the free edition is plenty. Medium teams that need directory sync or SSO: Professional, and get the quote in writing. Regulated or large: Enterprise for the compliance tooling. The catch: the price came off the website. You now have to talk to a salesperson to find out what the tier you need costs, and there is still no middle step between the free edition and a quoted annual contract. SSO moving down into Professional is a genuine improvement. Having to ask what Professional costs is not.

Chatwoot37.0k★

Chatwoot is an open source customer support and live-chat platform, a self-hosted alternative to Intercom and Zendesk. Customer messages come in from any channel, your team responds from one dashboard. Self-hostable with Rails and PostgreSQL. Used by startups and mid-size companies that want multi-channel support without Intercom's pricing. Self-hosted is free with no feature restrictions. Their cloud plan starts at $19/agent/mo and includes the same features plus managed hosting and support. The catch: Chatwoot is a support tool, not a product analytics or engagement platform. Intercom bundles marketing automation, product tours, and user tracking; Chatwoot doesn't. The self-hosted setup requires Ruby on Rails, PostgreSQL, Redis, and Sidekiq. That's a meaningful ops commitment. And the UI, while functional, lacks the polish of Intercom's refined experience. If you just need a chat widget and basic ticketing, it's great. If you need a full customer engagement platform, you'll outgrow it.

python-telegram-bot29.5k★

python-telegram-bot wraps Telegram's Bot API so you can build a bot without hand-rolling HTTP calls. It covers the whole API and adds higher-level helpers for handling commands, messages, and updates, so a working bot is a short script rather than a weekend. Mature, well-documented, LGPL-3.0, and free to install with pip. It is built entirely on asyncio, which is the modern way to do this and also the main thing to know going in: it is explicitly not thread-safe, so it does not slot cleanly into multi-threaded designs. For the overwhelming majority of bots, that is a non-issue, you write async handlers and move on. The library has almost no running cost and the maintainers are clear they are not chasing donations or a business model. Building anything on Telegram in Python, from a personal notification bot to a production service handling real traffic, this is the standard choice. Solo and team alike: free, no tiers, no gated features. The only reason to skip it is if you are not on Python, or you need a single one-off API call where the full library is more than you want. The catch is small but worth naming: asyncio-only and not thread-safe. Design around that from the start and you will never feel it. Bolt it onto a threaded app late and you will.

discord.js26.8k★

Discord.js is the most popular way to do it in JavaScript. It wraps Discord's API into a clean, event-driven library so you're writing bot logic instead of wrestling with WebSocket connections and rate limits. Fully free. Apache 2.0 license, no paid tier, no hosted service. You write your bot, host it wherever you want, and connect it to Discord. The library handles caching, sharding (splitting your bot across processes for large servers), slash commands, buttons, modals, and all the Discord interaction types. The catch: Discord's API changes frequently, and discord.js tracks it closely, which means major version upgrades can break your bot. The jump from v13 to v14 was painful for a lot of people. You also need somewhere to host the bot 24/7 (a $5/mo VPS works), and Discord's rate limits will bite you if you're not careful with how often your bot makes API calls. For Python devs, discord.py is the equivalent library.

Zulip25.9k★

Zulip fixes chat overload with threading that actually works. Every message belongs to a topic within a stream (channel). You can follow specific topics, mute others, and catch up on what matters without reading everything. It's how chat should have worked from the start. Apache 2.0. Full-featured: markdown, code blocks, integrations (GitHub, Jira, PagerDuty, 100+), search, file sharing, video calls, and mobile apps. The web app is fast. The threading model means you can be away for a day and catch up in 10 minutes by scanning topic titles. Zulip Cloud: free for small teams (up to 10K messages searchable). Standard at $6.67/user/month for full history and features. Self-hosting is free and unlimited. The catch: threading is Zulip's superpower and its adoption barrier. People used to Slack's flat channel model find it confusing at first. Convincing a team to switch from Slack requires everyone to learn a new interaction pattern. And the ecosystem is smaller: fewer third-party integrations, fewer bots, less community tooling than Slack.

LiveKit21.0k★

LiveKit started as open source WebRTC plumbing and now calls itself an end to end realtime stack for connecting humans and AI. The media server, the Agents framework, recording, egress and ingress, and SDKs for React, Swift, Kotlin, and Python are all Apache 2.0 with nothing held back. Self-hosting is free and expensive at the same time. Media processing burns CPU and bandwidth, so one server handles roughly 50 to 100 concurrent participants. This is not a $5/mo VPS. Add a TURN server, monitoring, and somebody who can explain why a call degraded. LiveKit Cloud finally publishes real numbers. Build is $0 with 1,000 agent session minutes and 5,000 WebRTC minutes. Ship is $50/mo, Scale is $500/mo with up to 600 concurrent agent sessions and SOC 2 reports, and SSO plus a support SLA require Enterprise. Solo and small teams should stay on Cloud; self-hosting pays off only at steady concurrency. The catch: the voice agent bill is not the media bill. Speech to text, the LLM, and text to speech each meter per minute on top of that, and they add up faster than the infrastructure. jitsi/jitsi-meet is the simpler answer when all you need is video rooms.

Lark CLI17.2k★

This official CLI covers everything. Messenger, Docs, Base, Sheets, Calendar, Mail, Tasks, Meetings, and more across 200+ commands with 19 pre-built AI Agent skills. The setup is fast: 3 steps from install to first API call, with one-click app creation and interactive login. Built in Go, MIT licensed. Every command has been tested with real AI agents, so the output formats and error handling are designed for machine consumption, not just human use. This is official from Lark Technologies, not a community wrapper. The catch: this is only useful if you're in the Lark/Feishu ecosystem. If your team uses Slack and Google Workspace, there's nothing here for you. And 'AI Agent acts under your identity' means the agent has your permissions. A hallucinating agent could send messages or modify documents as you. The security model requires careful scoping.

ZeroMQ11.0k★

ZeroMQ is a messaging library that acts like sockets on steroids, passing messages between processes reliably without a broker. No server to install, no broker to maintain. Just embed it in your app and start sending. ZeroMQ gives you patterns: pub/sub, request/reply, push/pull, and more. It handles reconnection, message queuing, and load balancing at the library level. Works across TCP, IPC, in-process, and multicast. Bindings exist for basically every language. Everything is free. MPL-2.0 license. No hosted version, no paid tier. It's a library you link into your code. Solo developers building distributed systems should seriously consider this before reaching for RabbitMQ or Kafka. Small teams benefit from the zero-ops angle: there's no broker to babysit. At scale, though, you'll eventually want the durability and management features that a real broker provides. The catch: no message persistence. If a consumer is down, messages can be lost. And "brokerless" means YOU are responsible for topology and failure handling.

centrifugo10.8k★

Centrifugo is a standalone server that handles all the WebSocket connections so your backend doesn't have to. Your app publishes messages to Centrifugo via HTTP API, and Centrifugo pushes them to connected clients. Language-agnostic: works with any backend. Apache 2.0. Supports WebSocket, SSE (Server-Sent Events), HTTP streaming, and GRPC. Built-in presence tracking (who's online), message history, and channel permissions. Handles millions of concurrent connections. Written in Go. The open source version is fully featured for most use cases. Centrifugo PRO adds features like push notifications (iOS/APNs, Android/FCM), analytics, SSO, real-time CEL-based channel permissions, and priority support. PRO pricing is custom. Contact them for quotes. Self-hosting: single binary, minimal config. Connect it to Redis or Nats for scaling across multiple nodes. Ops: moderate at scale (Redis cluster management), trivial for a single node. Solo and small teams: the open source version handles everything. Medium teams: still free unless you need push notifications or advanced analytics. The catch: adding a separate real-time server is another piece of infrastructure. If your framework already has decent WebSocket support (Phoenix, Rails Action Cable), you might not need it. Centrifugo shines when your backend is a REST API that wasn't built for real-time.

Typebot10.4k★

Typebot gives you a visual drag-and-drop builder for conversational forms, support intake, and onboarding flows. You connect blocks (text, input, logic, integrations) into a flow, embed it on your site, and collect responses. Like Typeform but for conversations, and you can self-host it. The builder supports conditional logic, variables, webhook integrations, OpenAI/Anthropic API calls, Google Sheets sync, and custom code blocks. Embed as a popup, bubble, or full-page chat. The cloud version is free for 200 chats/month. Pro is $39/mo for 10K chats. Self-hosting is free and unlimited. You run it with Docker and a Postgres database. The catch: the license is marked 'Other'. Check the actual terms before building a business on it. The self-hosted version requires you to manage your own infrastructure (Postgres, S3 for file uploads, SMTP for emails). And 200 free chats per month on cloud is tight. One moderately successful landing page burns through that in a week.

pixelfed7.1k★

Pixelfed is a federated photo sharing platform, basically Instagram without the algorithm or the ads. It connects to the fediverse through ActivityPub, so your users can follow and interact with accounts on Mastodon and other federated services. You own the data, you set the rules. Self-hosting runs on PHP with Laravel, requiring a web server, Redis, and a database (MySQL or PostgreSQL). The setup is a standard Laravel deployment, nothing exotic, but you are responsible for storage (photos eat disk space fast), background workers for federation, and keeping PHP updated. Completely free, no paid version. Solo developers or small communities can run an instance for their group. Larger communities need to plan for object storage (S3 or equivalent) and CDN distribution to handle media at scale. The catch: federation is both the feature and the complexity. ActivityPub brings interoperability but also moderation headaches from other instances. And you are competing with Instagram's network effect, which is the real mountain to climb.

netty-socketio7.0k★

Netty-socketio is a Java implementation of the Socket.IO server protocol built on Netty's high-performance networking. This lets your Java server speak Socket.IO to any Socket.IO client (JavaScript, Swift, Kotlin, etc.). You get namespaces, rooms, event-based messaging, and automatic transport fallback (WebSocket to long-polling). Netty underneath means it handles thousands of concurrent connections without breaking a sweat. Fully free. Apache 2.0 license. No paid tier. Java teams building real-time features who don't want to run a separate Node.js service just for Socket.IO will find this useful. It's been around since 2012, so it's proven. The catch: development has slowed significantly. It doesn't support the latest Socket.IO v4 protocol features. If you're starting fresh, consider whether you actually need Socket.IO compatibility or if plain WebSockets (with Spring WebSocket or Quarkus) would be simpler.

The Lounge6.3k★

The Lounge is a self-hosted IRC client that runs in your browser and stays connected 24/7. It's your own always-on IRC bouncer with a modern web interface. MIT license, TypeScript. It keeps you connected to IRC servers even when you close the browser tab. Message history persists across devices. The UI is clean and responsive. Works on desktop and mobile. Supports multiple IRC networks, file uploads, link previews, and push notifications. Fully free. No paid tier, no hosted version. The entire application is open source. Self-hosting is trivial. npm install, configure your IRC networks, access via browser. Runs on any VPS. A $5/mo server handles it with room to spare. Almost zero ongoing maintenance. Solo: perfect use case: always-on IRC for $5/mo. Small teams: works as a shared IRC gateway. Medium to large: if your team uses IRC (unlikely in 2026), this works, but most teams have moved to Slack, Discord, or Mattermost. The catch: this is an IRC client. IRC itself is a niche protocol in 2026. If your community or project isn't already on IRC, you don't need this. And compared to full team chat platforms, IRC lacks threads, reactions, integrations, and file sharing beyond basic uploads.

Plunk5.5k★

Plunk handles both transactional emails (welcome emails, order confirmations) and marketing emails (newsletters, drip campaigns) from one open source platform you control. Self-host it, connect your SMTP or AWS SES, and you've got email infrastructure without paying per-email fees to Mailchimp or SendGrid. The dashboard is clean. Manage contacts, build emails with a visual editor, set up automations, track opens and clicks. It's what you'd expect from a modern email platform, minus the monthly bill. AGPL-3.0, TypeScript. They also offer a hosted version. The catch: AGPL license means if you modify Plunk and offer it as a service, you must open source your changes. The project is relatively small compared to alternatives: fewer integrations, less battle-testing at scale. And email deliverability is hard. Self-hosting email means managing SPF, DKIM, DMARC, IP reputation, and bounce handling. Plunk makes the app layer easy but can't solve the deliverability puzzle for you.

Synapse4.6k★

Synapse is the reference Matrix homeserver, now maintained by Element. Matrix is to chat what email is to mail: federated and interoperable, so your server talks to every other Matrix server and nobody is locked inside one company's app. The core is free under AGPL v3, with encrypted rooms, spaces, voice and video, bots, and bridges into Slack, Discord, IRC and Telegram. Running it is real work. Python and Rust over Postgres, behind a reverse proxy, with DNS and federation verification to get right on day one. A few hundred users wants 4GB of RAM and climbs from there, and media storage grows without asking. Federation is the point of Matrix and also what makes moderation and performance harder. Solo: skip it unless running infrastructure is the fun part. Small teams: viable when someone actually wants the ops job, and Rocket.Chat or Zulip are lighter when nobody does. Medium teams: bridges and federation start earning their keep, so budget dedicated ops time. Large teams: price Element's hosted tiers before committing an engineer to this. The catch is subtler than the sales page suggests. Synapse relicensed from Apache 2.0 to AGPL v3 under Element, which also sells a commercial license. OIDC and SAML single sign-on both work on the free build, so do not let the pricing table talk you out of self-hosting. What Element actually gates is the packaged enterprise layer: LDAP directory sync, SCIM provisioning, auditing and retention tooling, and a separate homeserver called Synapse Pro. Community is pitched at 1 to 100 users and non-commercial use, with no support at all unless you subscribe.

WechatOnCloud3.8k★

WechatOnCloud runs WeChat on your own server and hands it to you through a browser. It containerizes the official, unmodified WeChat client behind a virtual display and KasmVNC, then wraps it in an admin panel so several people can reach the same session from any device. If you, or a family or small team, share a WeChat account and are tired of juggling clients across phones and laptops, this centralizes it. Self-hosting is the whole model, and the ops burden is moderate. You run it in Docker on a VPS or a NAS, it supports amd64 and arm64, and it gives you multi-instance management, role-based access for sub-accounts, one-click install and updates, and file and clipboard transfer between devices. There is real engineering here: soft-locking so two people don't fight over one session, PWA support for an app-like feel. This is niche by design and most useful in the WeChat-heavy world where one account often gets shared. Solo users who just want WeChat on a machine without the desktop client, and small teams managing a shared account, are the fit. It costs nothing to run beyond your own hardware. The catch is that you are running an unmodified proprietary client inside a container, which puts you in a gray area against WeChat's terms, and routing a personal messaging account through a shared server carries obvious account-safety and privacy risk. Useful and clever, but understand what you are signing up for before you put a real account behind it.

ergo3.3k★

Ergo is an IRC server that ships with the parts you used to bolt on yourself. A traditional setup means an ircd, a separate services package for nickname and channel registration, and a bouncer per user so nobody loses scrollback when they close the laptop. Ergo is one Go binary covering all three. MIT, free, no paid edition. Setup is a binary and a YAML file that rehashes at runtime, so most config changes never drop a connection. History sits in memory by default; point it at MySQL, PostgreSQL, or SQLite to survive restarts. TLS, SASL, bcrypt password hashing, and host cloaking are built in. A small VPS runs it fine. Solo and small private groups: cheapest self-hosted chat you can stand up in an evening. Teams already on IRC: the built-in bouncer alone justifies dropping ZNC. Anyone shopping for a Slack replacement wants mattermost/mattermost, zulip/zulip, or RocketChat/Rocket.Chat instead. The catch is IRC. No threads, no file uploads, no reactions you can count on, and what each person sees depends on the client they picked. Ergo modernizes the protocol as far as IRCv3 allows and stays honest about the ceiling. Watch the security releases: the August 2026 patch fixed an account verification bypass.

wecom-cli3.1k★

WeCom CLI drives WeChat Work from the terminal. Send messages, manage contacts, push approvals, and wire any of it into scripts or CI, with output shaped so an AI agent can consume it as easily as a person can read it. Official tool from the WeCom team, MIT licensed, free. There is nothing to host. Install the binary, authenticate with your WeCom app credentials, and you are issuing API calls. The design assumption is the interesting part: agent use is not an afterthought bolted on, it is the reason the output formatting exists, which is the same bet GitHub and Slack made with their own CLIs. Cost is not the deciding factor here, because there isn't one. The real sizing question is geography. WeCom is Tencent's enterprise messaging platform and it dominates in mainland China, so this is either indispensable to your company or completely irrelevant to it, with almost nothing in between. The catch follows from that. If your team is on Slack, Teams, or Discord, there is nothing here for you, and no amount of CLI polish changes it. Documentation and issues are Chinese-first as well. Well built for the audience it has, invisible to everyone else.

weixin-agent-sdk1.3k★

This SDK bridges that gap. Scan a QR code, install the plugin, and your AI agent is live in WeChat in about 2 minutes. It uses the Agent Client Protocol (ACP) standard, so if you already have an ACP-compatible agent, you can connect without writing code. Otherwise, you implement a simple Agent interface to hook up any backend. MIT licensed. Built in TypeScript. Adapted from Tencent's official @tencent-weixin/openclaw-weixin package. The catch: this is not an official Tencent project. It's adapted from official code for 'learning and communication purposes.' That's a legal disclaimer worth taking seriously if you're building a business on it. WeChat's platform policies around bots are strict, and unofficial integrations can get your account flagged. The SDK targets China's WeChat ecosystem specifically. If your users aren't on WeChat, this isn't for you.