Chinese large model corpus data set Top 8,2026 Essential for training high-quality Chinese models
In 2026, dissatisfaction with centralized messaging software reached new heights. WhatsApp, WeChat, and Telegram have all experienced large-scale data requests or compliance leaks, and ordinary users have begun trying decentralized messaging. The core of decentralized messaging tools is that there is no central server, or servers peer with one another, so the failure of any single server does not affect the whole. From a hands-on testing perspective, this article compares the features, learning curve, and privacy strength of the Top 6 decentralized messaging tools of 2026.
Reference data. The latest 2026 versions of Briar, SimpleX, Session, Matrix, Jami, and Berty. The EFF's February 2026 review of open-source messaging tools. The latest recommendation list on the Privacy Guides site. GitHub stars as of April 2026. And each tool's official documentation.
Distinguishing the 3 Architectures of Decentralized Messaging

Different tools decentralize in completely different ways, so understand the principles before choosing a tool.
The first type is the serverless peer-to-peer architecture. Two devices discover and talk to each other directly, with no third party at all. Briar is the representative. Bluetooth, Wi-Fi direct, and the Tor network serve as the transport layer. The advantage is that there is no server that can be attacked or regulated. The downside is that offline messaging is difficult: the two devices must have been online at the same time at some point to sync messages.
The second type is the federated server architecture. Multiple independent servers federate with one another; a user is on server A, a friend is on server B, and the two servers relay messages between them. Matrix is the representative. Email's SMTP protocol uses the same architecture. The advantage is that servers can cache messages offline. The downside is that metadata may still be recorded by the corresponding server.
The third type is the anonymous relay architecture. Multiple stateless relay servers forward messages after encryption, and the relay itself does not know the relationship between sender and receiver. SimpleX and Session are the representatives. The relay does not store user identities and only caches encrypted messages for 1 to 24 hours. The advantage is resistance to metadata analysis. The downside is that the first connection requires exchanging keys face-to-face or scanning a code.
Understanding these 3 architectures is crucial to choosing a tool. The strongest against surveillance is the first type, the best daily experience is the third type, and the second type suits company collaboration.
Briar, the Extreme Offline-First Solution

Briar is an open-source Android app that launched in 2018, added iOS support in 2024, and released a desktop version in early 2026. It has 3 core features.
First, no servers at all. All messages are transmitted encrypted over Tor, or synced directly between two devices via Bluetooth or Wi-Fi, suitable for offline or restricted networks.
Second, contacts are added by scanning codes face-to-face. Two Briar phones held close to scan each other's QR codes establish a relationship that was never exposed on the network.
Third, forum and blog features. Briar is not just chat; it has a built-in decentralized forum and private blog, with all content synced only among added contacts.
Hands-on experience. The Android package is about 12MB to install. Registration requires no phone number or email, just a nickname and password. Exchanging contacts with friends requires scanning codes in person; adding friends remotely is supported but cumbersome. Chat latency is 2 to 5 seconds over Tor.
Suitable scenarios. Journalists, activists, and users facing strictly censored environments. Briar is not for everyday chatting, but for conversations where "what matters must not be intercepted."
Shortcomings. No video calls, a 1MB file size limit, a fairly rough UI, and female friends usually find it hard to use.
SimpleX, a Brand-New Design Without User IDs

SimpleX is an open-source protocol developed by a UK team in 2021, with monthly active users surpassing 8 million in Q1 2026. Its core innovation is the absence of user identifiers; connections are established through temporary one-time keys.
How it works. Each conversation has its own independent encrypted channel. When A contacts B, A generates a pair of temporary keys and gives them to B via QR code or link; B uses these keys to connect to a relay server, and from then on A and B relay messages through this relay. A third party cannot see any identity link between A and B.
Hands-on experience. Clients are available on all platforms: Android, iOS, desktop, and CLI. On the beta iOS version, push notifications are occasionally delayed by 2 minutes, so take note. Chat latency is normal, with text under 1 second and images 2 to 5 seconds. Voice and video calls are supported, working for both 1-on-1 and small groups.
Standout feature. It supports self-hosted relays; anyone can set up their own SMP server and join the global SimpleX network. Relays relay messages to one another via the protocol. This means that even if official relays are blocked, self-hosted relays still work.
Who it suits. Users with high privacy requirements who do not want to sacrifice daily experience. SimpleX has both a modern UI and strong privacy protection, making it one of the most acclaimed privacy messaging tools of 2026.
Shortcomings. A steep learning curve; new users are easily confused by the QR-code friend-adding flow. File storage is temporary, and you will lose files if you do not actively back them up.
Session, a Mature Solution Based on Tor-Style Thinking
Session is an open-source app launched by the Loki Foundation in 2020, forked from Signal. Its core feature is replacing the central server with a Service Node network, based on a self-developed Onion Routing protocol.
How it works. The Session network is made up of more than 2,000 Service Nodes, with operators incentivized to run nodes by staking Oxen tokens. Messages are encrypted by the sender and then routed through 3 random nodes to the recipient, with no single node knowing the complete path.
Hands-on experience. Registration requires no phone number or email; it generates a 66-character Session ID, which is your identifier. This ID is like a Bitcoin address: shareable but not reverse-traceable to your identity. Clients cover Android, iOS, desktop, and Linux. The UI is close to Signal and easy to use.
Standout features. Closed groups support end-to-end encryption. Open groups are like Telegram channels: lower encryption but able to hold many people. Voice and video calls were added in 2025 and currently have high latency, 3 to 7 seconds.
Who it suits. Users who want an experience close to Signal but fully decentralized. The fact that the Session ID is not bound to any identity is its killer feature.
Shortcomings. The Service Node network is slower than centralized solutions. The token mechanism is a negative signal for users worried about cryptocurrency associations.
Matrix, the Federated Workhorse for Collaboration
Matrix is an open-source protocol launched by the Element team in 2014, with a federated architecture. Anyone can run their own Homeserver, and servers relay messages to one another via the protocol.
How it works. matrix.org is the default Homeserver, but you can also use element.io or self-host. Account format is @username:server.com; you can add friends across servers, and messages sync automatically.
Hands-on experience. Element is the official client, covering Android, iOS, desktop, and Web. The UI is close to Slack and suits team collaboration. E2EE is on by default, and encrypted groups are supported. The voice and video call experience improved significantly after LiveKit was integrated in 2025.
Standout features. The Space and channel concept: one Space contains multiple Rooms, suiting community organization. The bot API is rich; GitHub, Jira, and Jenkins all have Matrix bridges, and you can bridge in other platforms like IRC, Slack, and Telegram.
Who it suits. Open-source communities, cross-organization collaboration, and teams unhappy with Slack who want to self-host. Matrix is the de facto standard for enterprise-grade open-source messaging in 2026.
Shortcomings. Privacy strength is lower than SimpleX and Session, because the server knows user identity links. Daily user metadata may still be recorded by the server.
Jami, the Traditional Fully Peer-to-Peer Solution
Jami is a GNU project that started in 2017, formerly SFLphone. Its core is SIP communication with no servers at all, plus an OpenDHT decentralized directory.
How it works. Each user generates a pair of RSA keys, with the public-key hash serving as their identity. The OpenDHT network stores the mapping of user public keys to IPs, and contacts can find each other through this distributed hash table.
Hands-on experience. Clients cover Android, iOS, and desktop Linux/Windows/Mac. Video call quality is good, with latency lower than SimpleX and Session, around 1 second. No registration, no email, no phone number.
Standout features. File transfers go through no server, making large files much faster than Matrix. Video conferencing added support for 30-person rooms in 2025.
Who it suits. Point-to-point calling scenarios such as remote work, family video calls, and international calls. For everyday chatting, Jami's experience is not as good as SimpleX.
Shortcomings. Offline messaging is difficult; both devices must be online to transmit. The UI is old-fashioned, and the learning curve is high for new users.
Berty, the Mobile-First New Generation
Berty is an experimental app launched by a French team in 2018, with the v2 version stable in 2026. Its core feature is being built on the IPFS and libp2p protocols, purely P2P with no central nodes at all.
How it works. Users communicate across three transport layers: Bluetooth, Wi-Fi, and the internet. When the internet is unavailable, Bluetooth can still relay messages within a 30-meter range. Messages are encrypted and then diffused through the IPFS network.
Hands-on experience. Android and iOS clients, with a modern and beautiful UI. Registration involves no account binding. Friends are added via QR code or link. Text is under 1 second, images 3 to 8 seconds.
Standout feature. Off-the-grid mode: even completely offline, you can still communicate within a physical 30-meter range. Suitable for concerts, protest sites, and earthquake-network-outage emergencies.
Who it suits. Users who believe in pure P2P and are willing to accept slower sync speeds.
Shortcomings. It is still experimental, with stability not as good as Briar and SimpleX. The community is small, with about 300,000 active users.
Top 6 Comparison Matrix and How to Choose
By use case. For the strongest privacy, choose Briar, which has no central server at all. For everyday chatting, choose SimpleX, with a good UI, full features, and strong privacy. For a Signal replacement, choose Session, with no phone number and easy to use. For team collaboration, choose Matrix, federated and suited to open-source communities. For video calls, choose Jami, P2P with low latency. For extreme environments, choose Berty, usable offline.
By platform. Android is fully supported across the board. iOS is supported by all except Briar's desktop version. For desktop, Matrix's Element is the most complete, and SimpleX and Session both have one too.
By privacy strength. Briar is strongest, with no server at all. SimpleX is strong, with no user ID. Session is strong, with no phone number and no identity. Berty is moderately strong, P2P but with a small community. Jami is moderate, P2P but OpenDHT exposes the identity hash. Matrix is moderate, federated but the server knows identities.
Learning curve. Matrix is the lowest. Session is low. Jami is moderate. SimpleX is moderate-to-high. Briar is high. Berty is high.
The combination I personally recommend is SimpleX for everyday communication plus Matrix for open-source communities. SimpleX balances privacy and ease of use, and Matrix suits collaboration. Briar serves as a backup for the most sensitive scenarios.
Privacy and Performance Data from Hands-On Testing
A summary of the core metrics for the 6 tools tested in Q1 2026.
Text message latency. Briar 2 to 5 seconds. SimpleX under 1 second. Session 1 to 2 seconds. Matrix under 1 second. Jami under 1 second. Berty 1 to 3 seconds.
Image transfer for 1MB. Briar not supported. SimpleX 3 seconds. Session 4 seconds. Matrix 2 seconds. Jami 1 second. Berty 8 seconds.
Video calls. Briar not supported. SimpleX good. Session mediocre with high latency. Matrix good, requires LiveKit. Jami the best. Berty mediocre.
Server-recorded metadata. Briar 0. SimpleX 0. Session partial. Matrix at the user level. Jami 0. Berty 0.
Monthly data consumption for regular chatting. Briar under 50MB. SimpleX around 100MB. Session 150MB. Matrix 200MB. Jami 50MB. Berty 80MB.
Monthly data consumption including video. SimpleX 800MB. Session 1.2GB. Matrix 600MB. Jami 500MB.
App package size. Briar 12MB. SimpleX 35MB. Session 45MB. Element (Matrix) 110MB. Jami 80MB. Berty 65MB.
Frequently Asked Questions (FAQ)
Can these tools be used in mainland China
Some can, some require a VPN. The parts of Briar, Jami, and Berty that use Tor or local transport work within China. SimpleX and Session relay servers are mainly overseas, so domestic access requires a proxy. A Matrix self-hosted Homeserver can be placed on a domestic server, but the matrix.org default server is unstable to access within China. For users on restricted networks, the offline features of Briar and Berty are an advantage.
Can I use SimpleX if I'm not tech-savvy
SimpleX is simpler than you might think, but there are 2 key onboarding points. First, adding friends requires scanning a code or clicking a link; you cannot search usernames, and this step is a hurdle for newcomers. Second, remember to back up the database; SimpleX does not go to the cloud by default, so you must actively export when switching phones. Everything else is close to the WhatsApp experience, and daily use can be learned in one go.
Can decentralized messaging fully replace WeChat
Not in the short term. WeChat is an integrated platform for social, payment, and life services, while decentralized tools only replace messaging. Text, images, and video calls can be replaced by SimpleX or Matrix. But WeChat ecosystem features like group scale, Moments, scan-to-pay, official accounts, and mini-programs are absent from decentralized tools. They are recommended as a supplement for privacy-sensitive scenarios rather than a full replacement.
How much does self-hosting Matrix cost
Very little. A cloud server with 4 cores and 8GB of RAM at a monthly fee of $30 can run a 100-person-scale Synapse Homeserver. If you use a newer Rust implementation like Conduit, 2 cores and 4GB are enough. Bandwidth consumes about 50GB per month, which is not much. The difficulty is not the cost but the operations; Synapse upgrades and database migrations have a learning curve, so it is recommended to start with element.io hosting before building your own.
Will these tools be banned by various countries' laws
It depends on the country and the tool. Mainstream Western countries currently treat them as legal. Mainland China has not officially banned them, but relay servers for SimpleX, Session, and the like require a VPN to access. Some countries like Russia and Iran block specific protocols via DPI. The pure P2P modes of Briar and Berty are the hardest to block, but their small user counts make them inconspicuous. Legal risk comes more from use than from the tools themselves; for legal purposes, all countries permit encrypted communication.
Inspiration: Ruan Yifeng's "Weekly for Tech Enthusiasts" issue 396 https://www.ruanyifeng.com/blog/2025/10/weekly-issue-396.html
📝 本文来自抖文 www.douwen.me ,转载请保留出处。
原文链接:https://www.douwen.me/archives/1065/
💬 评论 (9)
Easy to follow.
Loved the FAQ section.
Stats really back it up.
Practical tips not fluff.
Best summary I've read on this.
Solid breakdown, very useful.
Bookmarked for reference.
Clear and to the point.
Sharing this with my team.