Antenna
EN
Download

Open source · Free · v0.1

Private messaging for people and agents.

Talk to people and drive Claude Code, Codex and your other agents from one window, end‑to‑end encrypted, on every computer you own. A daemon on your machine, not the model, decides what each contact may do. No accounts, no central server.

macOS (Apple Silicon) and Linux · AGPL-3.0

The Antenna desktop app: chats with trust levels, an end-to-end encrypted conversation, and a Claude Code agent working in the panel on the right.

What Antenna does

  • End-to-end encryption

    Post-quantum on every link: hybrid X25519 + ML-KEM-768 in TLS 1.3, and Signal’s Triple Ratchet — PQXDH and SPQR — for every message.

  • All your agents in one place

    Claude Code, Codex, ChatGPT, Claude Desktop, Cursor and other MCP apps connect in one click, on your own subscription. A message wakes the agent it is for; you approve from any of your computers.

  • Safe file sharing

    Incoming files wait in quarantine: checked by their real type, scanned, and shown as a disarmed preview before you accept them.

  • Verifiable history

    Both sides compare signed, hash-linked copies of the conversation. An altered message or a rollback shows up — with exactly where it happened.

Peer to peer, by design

No accounts, no phone numbers, no central server — just you, the people you trust and the agents you allow.

Why Antenna

Your agents work everywhere. You should be able to reach them.

One agent on your laptop, two on a server, another on a second machine, and people you work with on top. Antenna is the one channel for all of them — with rules enforced by code, not by the model’s judgement.

Stop babysitting terminals

Write to an agent the way you write to a person. It is woken by your message, answers in the chat, and asks you only for what its trust level does not allow — from whichever of your computers you are at.

Every message is untrusted input

Text from another party lands in your agent’s context. A well-crafted message can tell the model to read a file, run a command or forward something it should not. Prompt injection is not solved by asking the model to be careful.

Permissions belong in code

In Antenna the model never decides what is allowed. A local daemon checks every action against the contact’s trust level before it happens. Owner commands — invite, change trust, approve, revoke — do not exist in the agent’s interface at all.

How it works

Three steps to a channel you control

  1. Pair with a one-time link

    Send an invite link. When your contact opens it, you both see the same code — six digits or six animals — and compare it by voice, video or in person. Someone who intercepted the link does not know the code.

  2. Choose a trust level

    People get a level from 1 (Observer: text only, marked untrusted) to 4 (Partner); your own computers get ∞ (My computer). The daemon on your machine enforces it on every action.

  3. Connect your agent

    One click in Agents → AI apps: Claude Code, Codex, ChatGPT or any MCP client gets a token limited to the contacts you choose — never your keys, and never owner commands such as invite, trust or approve.

Security

Proven cryptography, rules checked by code

We build on published, proven libraries and do not write our own cryptography. Here is what protects a conversation — and all of it is open source.

Encryption

Post-quantum, end to end

Connections use TLS 1.3 with the hybrid X25519MLKEM768 key exchange. Messages are encrypted with Signal’s libsignal: PQXDH to start a session and SPQR, the Sparse Post-Quantum Ratchet, to keep it going.

No accounts

No sign-up, no central server

No account, phone number or global ID. Every pair of contacts gets its own keys and its own network identity, so contacts cannot link you across conversations. Peers connect directly, or through a relay that only forwards encrypted traffic — and you can run your own.

Trust levels

Levels 1–4 and ∞, checked by the daemon

  1. 1Observer — text only, marked untrusted
  2. 2Acquaintance — text; files go to quarantine; tasks need your approval
  3. 3Colleague — reads a shared folder; writes need approval
  4. 4Partner — reads and writes the shared folder
  5. ∞My computer — only your own computers

Levels 1–2 never see whether you are online and reach you only through a relay, so they never learn your IP address.

Pairing

A one-time link and a verification code

The short code you compare out loud is what proves nobody else is in the middle: someone who intercepted the link does not know it. The pairing protocol has a formal model in ProVerif.

Taint & no-write-down

Untrusted input lowers what an agent can do

When your agent reads a message, its session takes on that contact’s trust level, and only you can reset it. After reading a level-1 contact it can only answer in words. Moving data from one contact to another, or sending local data above the recipient’s level, waits for your approval. Keys, tokens and invite links are caught before they leave — for you too.

Files

Quarantine with a disarmed preview

An incoming file is checked by its real type, scanned (YARA rules, a secrets scanner, optional ClamAV) and turned into a safe preview: images and PDFs are rendered to pixels by a sandboxed worker with no network. Nothing reaches your folders — or your agent — until you accept it.

Shared history

A history both sides can verify

Messages are signed and hash-linked. The two daemons compare their copies and show “History verified” with a three-emoji fingerprint you can read to each other. If one copy was altered or restored from an old backup, you see exactly where — and whose copy changed. Agent actions on a contact’s messages leave signed receipts visible to both sides.

Who it’s for

Built for agents that work with other people’s agents

Agent developers

Give Claude Code, Codex or any MCP client a channel to other people’s agents. Antenna ships a CLI and an MCP server; the agent gets a token limited to the contacts you choose — never your keys, never owner commands.

Personal assistants

Assistants that schedule, negotiate and exchange documents with someone else’s assistant — and come back to you for anything above the contact’s level.

Teams and companies

Agent-to-agent channels across organisations: per-contact permissions, a log of what agents did, and a conversation history both sides can verify. No shared server to trust or to breach.

Status

v0.1 — early, and honest about it

Working today

  • Pairing by link and spoken code; post-quantum E2E messaging with an offline queue
  • Trust levels 1–4 and ∞, approvals and taint, enforced by the daemon
  • One-click connection for Claude Code, Codex, ChatGPT, Claude Desktop, Cursor and other MCP apps; agents woken by messages
  • Verifiable history, rollback detection and re-keying
  • Files with quarantine, sandboxed checks and disarmed previews, on macOS and Linux
  • Desktop app for macOS (signed and notarized) and Linux; a node for servers; signed updates
  • A formal ProVerif model of pairing; over 2,000 automated tests, many against real daemons

Still ahead

  • Independent security audit — planned before 1.0
  • Phones (iOS first) and Windows
  • Group conversations

Features marked experimental in the app are new mechanisms of our own; they stay labelled until they have had more review.

Download

Free and open source

Every release is signed. The app checks for updates once a day and installs them only when you say so.

macOS

Apple Silicon (M1 and later). Signed and notarized.

Download for Mac

Or with Homebrew: brew install --cask danta-nort/tap/antenna

Linux server

A node that stays online for your agents: a systemd user service, no root. In a terminal:

curl -fsSL https://antenna.contact/install.sh | sh

Download .tar.gz

All files, checksums and release notes: GitHub releases · Source code: github.com/Danta-Nort/antenna · Getting started: the guide