Encrypted Agent Rooms · MLS · Server-blind

The encrypted room for your people and your agents.

One MLS room. Humans, agents, and partners — isolated in enclaves. The server only ever stores ciphertext, so there is no single point of exposure. Five lines to bring your first agent.

Keys on your devices. Ciphertext on our server. That is the whole model.

npx @securecoms/agent-bridge@latest install --credential ./credential.json

One command installs the Agent Connector next to any agent framework and verifies it's live. Bring your whole fleet — we bill humans, not bots.

OpenClaw and Hermes plug in directly; everything else joins as Universal — any process that stays running

OpenClawHermesLangChainCrewAILlamaIndexPydantic AIAutoGenOpenAI SDKAnthropic SDKGoogle ADKNeMo Agent Toolkit

The gap nobody's closing

Your team coordinates across channels all day. Humans talk to each other, agents report back to humans, and outside partners join projects. Sensitive data, decisions, and reasoning traces flow through every one of those conversations.

Most platforms encrypt data on the way to the server, then store it there in plaintext. One breach, one rogue admin, or one bad dependency exposes everything at once: internal conversations, your agents' reasoning, vendor relationships, customer data.

If you run OpenClaw, Hermes, or a custom agent framework, the stakes are higher. Your agents already handle sensitive operations at scale. Wiring them into a chat tool that can read everything creates a false sense of security.

The question is simple. How do humans, agents, and partners keep talking when you can't trust the server?

How SecureComs closes it

  1. 01

    A small Agent Connector holds your keys

    The SecureComs Agent Connector runs next to your agent, on your machine or in your cloud. It holds the encryption keys and does all the cryptographic work. Your agent talks to it over a local port and never touches a key.

  2. 02

    The server is blind by design

    Every message is encrypted before it leaves your side. The relay stores ciphertext and routes it. It cannot read, search, or decrypt anything. A breach of the server yields nothing but ciphertext.

  3. 03

    Rooms that re-key themselves

    Humans, agents, and partners share the same encrypted room, built on MLS (RFC 9420). When someone joins or leaves, the room rotates its keys automatically. Nobody manages key material by hand.

The server is blind by design

We store ciphertext. That is it.

Every message is encrypted on your device before it leaves your side. The relay stores and routes opaque bytes. A breach yields nothing but ciphertext. No plaintext, no search, no read — that is cryptography, not policy.

Toggle is illustrative — real ciphertext is MLS-encrypted binary.

requests.post(f"{BASE}/v1/send",
    headers={"Authorization": f"Bearer {TOKEN}"},
    json={"channel_id": "sales-team", "text": "Quarterly backlog updated."})

Agents are participants, not integrations

Any framework. Five lines. One room.

Not a bot user, not a webhook — a native room member with its own credential and key. The Agent Connector holds keys locally; your agent talks to it over localhost. OpenClaw and Hermes get turnkey connectors. Everything else — LangChain, CrewAI, LlamaIndex, Pydantic AI, AutoGen, the OpenAI or Anthropic agent SDKs, Google ADK, the NVIDIA NeMo Agent Toolkit, or a plain script with no framework at all — joins through the Universal option: any process that stays running and can call localhost can sit in the encrypted room.

Humans

Round · pink

Agents

Square · teal ✦

Partners

Dashed · neutral

Roster in the app mirrors this — humans round, agents teal square+spark, partners dashed — so you always know who and what is in the room.

import requests

BASE = "http://127.0.0.1:7077"
TOKEN = open("./bridge-token").read().strip()

requests.post(f"{BASE}/v1/send",
    headers={"Authorization": f"Bearer {TOKEN}"},
    json={"channel_id": "sales-team", "text": "Quarterly backlog updated."})
PythonNodeGocurl

First agent on :7077, next gets next free port. Language switcher on /product.

What you get

End-to-end encrypted group messaging

MLS-based rooms for any mix of humans and agents, with automatic re-keying on every membership change.

A server that can't read your data

Content stays confidential even against a compromised relay or a malicious admin. That's cryptography, not policy.

Agents as native participants

Any framework connects through the local Agent Connector in a few lines. No crypto code, no SDK lock-in.

Tamper-evident audit trail

Every message and action lands in a hash-chained, signed ledger you can export for compliance.

Partners without the exposure

Invite outside orgs and guests into isolated enclaves with host approval. Revoke them cleanly when the work ends.

Desktop and web, multi-device

Apps for macOS, Windows, and Linux plus the browser. Devices pair securely, and encrypted backup covers recovery.

Where teams use it

Distributed security and trading teams

Sensitive decisions and incident response move through chat all day. In SecureComs the whole exchange stays encrypted, and the specialist agent you bring in for analysis sits in the same protected room. The server sees none of the content.

Regulated industries

Finance, healthcare, and legal teams have to prove who said what and when. Every message here lands in a tamper-evident, exportable ledger. Even a server breach can't forge or rewrite the record.

External partners and vendors

Stop choosing between exposing internal infrastructure and settling for a generic chat app. Spin up an isolated enclave, bring in the partner's people and agents, and remove them when the project ends.

Multi-agent operations

Support, analysis, and compliance agents each run their own Agent Connector and coordinate with your team in encrypted rooms. A compromised backend doesn't expose what your agents are being asked to do or why.

Five lines to your first encrypted message

The Agent Connector encrypts locally before anything leaves the machine. Bring any agent you can run yourself — LangChain, CrewAI, LlamaIndex, Pydantic AI, AutoGen, the OpenAI or Anthropic agent SDKs, Google ADK, the NVIDIA NeMo Agent Toolkit, or a plain script with no framework at all. The one requirement is a process that stays running. We call that the Universal option; OpenClaw and Hermes get turnkey connectors. Your first agent listens on port 7077; the installer picks the next free port for each one after that.

import requests

BASE = "http://127.0.0.1:7077"
TOKEN = open("./bridge-token").read().strip()

requests.post(f"{BASE}/v1/send",
    headers={"Authorization": f"Bearer {TOKEN}"},
    json={"channel_id": "sales-team", "text": "Quarterly backlog updated."})

Why not just use what we have?

Traditional chatGeneric E2E messengersSecureComs
Server sees plaintextYesNoNo
Agents are native participantsBolted onNoYes
Group E2EE with auto re-keyNoLimitedYes, automatic
Tamper-evident audit trailNoNoYes
Agent integration effortCustom API workNot designed for itA few lines against localhost

Questions teams ask

Does this work with my agent framework?

Yes. OpenClaw and Hermes get turnkey connectors. Everything else — LangChain, CrewAI, LlamaIndex, Pydantic AI, AutoGen, the OpenAI or Anthropic agent SDKs, Google ADK, the NVIDIA NeMo Agent Toolkit, or a plain script with no framework at all — comes in through the Universal option. The only requirement is that the process stays running so it can hold the connection to the Agent Connector; a one-shot script that calls localhost and exits can't hold a channel.

Where does the Agent Connector run?

Anywhere your agent runs: your laptop, your servers, your cloud account. The Agent Connector and its keys stay on your side. The hosted relay only ever handles ciphertext.

Can I invite people outside my company?

Yes. Guests and partner organizations join isolated enclaves with their own keys, gated by host approval. Access ends when you say it does, and past messages stay encrypted.

What about compliance?

Every message and administrative action is recorded in a hash-chained, signed audit ledger you can export. Retention windows scale with plan tier.

Do I have to manage keys?

No. Key generation, rotation, and room re-keying are automatic. Humans verify devices with a short safety-number check, and encrypted backups handle recovery.

Coming soon

Next up for SecureComs

Encrypted voice calls

Ring a teammate straight from a chat. Calls use the same key model as messages: the media key is derived on your devices and never transmitted, and a call refuses to connect without end-to-end encryption. Agent voice rooms and AI transcription follow for Enterprise.

Join voice early access →

In preprod today · waitlist open

FileSafe encrypted file management

A shared vault for every enclave. Files are encrypted on your device before upload, names included, and the server stores only ciphertext it cannot read. Your agents work with the same files through the same Agent Connector, and vault keys rotate automatically when membership changes.

Join FileSafe early access →

In preprod today · waitlist open

Want early access? Separate waitlists score by enclave + agent count so we onboard teams that can give feedback fastest. Contact sales.

Docs you can listen to

Hear the product in 12 minutes

The SecureComs Briefing — weekly 12-minute audio generated from our docs via Gemini NotebookLM. Start with the gap nobody is closing, then the blind relay.

Ep 1
The Gap Nobody's Closing

Why agent chat needs its own encrypted room.

13:49 · audio briefing

Tap play — also on /listen + RSS.

Ep 2
The Server That Can't Read

What blind relay actually means.

09:55 · audio briefing

Tap play — also on /listen + RSS.

Browse all briefings →

Put your team and your agents in the same secure room