All your chats.
One
epicenter.
Epicenter pulls your messaging apps, email, and your whole message history into one private inbox that runs on your own server. Bring your agents too. They only get what you hand them.
Chaos up there.
Calm down here.
Every account, every thread, and every person in one quiet workspace. It is built for reading and replying, not for watching badges pile up.
Chats
Chats across connected platforms.
1
All conversations
6 chats across 6 accounts
Field NotesThe draft is up for review in #writing
Maya Chen
Signal, personal account- Every account gets a seat in the dock.
- It only says “Delivered” when the network does.
- One person, many apps, one name.
- After a restore, sending waits for you.
Bring the whole group chat.
Messaging apps connect through open-source bridges. Every chat lands in the same place, tagged with where it came from.
- Signal
- Telegram
Slack- Messenger
- X
- and more
Email gets its own room.
Email isn’t squeezed in between group chats. It has its own inbox with categories, labels, stars, and keyboard shortcuts, and it lives in the same private archive, so one search finds both.
Inbox
24 shown
Everyone, once.
One list of people instead of a handle per app. Each contact brings together their accounts across networks, the chats you share, and where they work, so a look-alike name never gets mixed up with someone else.
22384164
Maya ChenProduct designer, @mayachen
Person
Field Notes
31248
Lisbon
Jonas WeberLandlord
Person
Weber Immobilien
23
Munich
1.2k64
Remote
Ana Ruiz@anaruiz
Person
Studio Norte
20412
Madrid
Kai Okafor@kai
Person
—
889
Berlin
Lena ParkSpeaker, @lenapark
Person
Conf Co
146
Seoul
See who you actually talk to.
Your whole network across platforms, ranked by the messages you really send and receive. The closest people sit nearest the middle.
Network
Your full network across platforms. Closest contacts are ranked by direct messages sent and received.
248 contacts, 15 shown
- #1
Maya Chen1,204 messages, 640 sent, 564 received - #2
Ana Ruiz860 messages, 402 sent, 458 received - #3
Kai Okafor512 messages, 251 sent, 261 received - #4FNField Notes433 messages, 119 sent, 314 received
- #5
Lena Park210 messages, 98 sent, 112 received
Meet Epicenter Agent.
An assistant that lives in your inbox. It can catch you up on a busy group chat, find something someone sent you months ago, or draft a reply for you to check. You pick the models it runs on, and you decide which conversations each one is allowed to read.
- What did I miss in Field Notes?
- When does Maya land?
- Find the lease Jonas sent me
Models
- Qwen 3.8 27BOpen weights, running on your Mac Local
- GLM 5.3Open weights, on your own server Self-hosted
- ClaudeAnthropic API Cloud
- GPTOpenAI API Cloud
Who can read what
- FamilySignal group Local only
-
Field NotesSlack channel
LocalSelf-hostedCloud
- Lease renewalEmail thread LocalSelf-hosted
- Bank alertsEmail Nobody
Any model you like.
Open-weight, private, self-hosted, or cloud. Mix them, and switch whenever you want.
Your rules, per conversation.
Each model only sees the chats you allow. A conversation you keep local never leaves your machines.
Same guardrails as any agent.
It works through a scoped grant, sending waits for your OK, and incoming messages stay data.
Bring your own agent.
Epicenter speaks MCP and a plain HTTP API, so the agents you already use can plug straight in. Each one gets a grant you control.
ChatGPT
Claude
Grok Bot
Muse
Hermes AgentCursor
OpenCode
OpenClaw
Instinct
…or anything else that can use MCP tools or call an HTTP API.
Its own key, its own lane.
Every agent gets a grant: which accounts, which conversations, which actions. Nothing more.
It asks before it sends.
Sending needs your OK by default, tied to the exact recipient and words. Change the draft and it asks again.
Prompt injection is just a message.
Incoming messages are treated as data, never as instructions. “Ignore previous instructions” in a group chat stays a joke.
- MCP
- REST + OpenAPI 3.1
- Webhooks
- TypeScript SDK
Maya asked if dinner’s still on. Want me to tell her you’re running ten minutes late?
Needs your approval
To Maya Chen on Signal
Running ten minutes late. Order without me!
Earthquake-proof.
A backup only counts once it restores, so recovery is a feature here, not a footnote.
Backups
Encrypted repositories, job history, and staged recovery.
Restore extracts into an isolated server directory with sending suspended. Nothing replaces this workspace until you say so.
-
Back it up
Your Postgres database and media are captured into an encrypted, Restic-compatible repository you control.
-
Take it anywhere
Export everything as an age-encrypted bundle, separate from full server recovery.
-
Restore, calmly
Restores are checked before they go live, and sending stays off until you switch it back on.
Works on its own. Even better with Portal.
Epicenter is the communications layer of Portal, the open context platform for your life. Let Portal hold your contacts and share context with the agents you trust.
Questions, answered.
Is Epicenter ready to use?
Not quite. It’s pre-release. The Rust server runs as a production candidate alongside the system it replaces, and each network, live Portal sync, and full archive parity switches on only after it passes its acceptance check.
Where do my messages live?
On your server, in a PostgreSQL database with your media stored next to it. Epicenter is self-hosted, so your messages, contacts, and backups stay on infrastructure you control.
Which apps can I connect?
The first release has connector profiles for WhatsApp, Signal, Telegram, Slack, Messenger, Instagram, LinkedIn, X, and email. Sign-in works differently on each network, and each one shows its real status.
Which agents work with it?
If your agent can use MCP tools or call an HTTP API, it can connect. Each agent gets its own grant, and sending waits for your approval unless you decide otherwise.
What is Epicenter Agent?
An assistant built into your inbox. It can catch you up, find things, and draft replies for you to check. You choose which models it runs on (open-weight, private, self-hosted, or cloud) and which conversations each model can read, and it follows the same grants and approval rules as any other agent. Under the hood it’s built with nanocode, an open-source SDK for embedded agents.
Is it open source?
Yes. The core is AGPL-3.0-or-later. The TypeScript SDK and the OpenAPI description are Apache-2.0.
Is this made by WhatsApp, Signal, OpenAI, or friends?
No. Epicenter isn’t affiliated with or endorsed by any of the networks or agents on this page. Their names and logos belong to their owners and are shown only to say what you can connect.


