Conch is a native Android client that lets you SSH into your own servers and drive AI coding agents — Claude Code, OpenAI Codex, and Google Gemini — that you've installed there.

It is a terminal in your pocket, not a hosted AI service. There are no accounts to create, no quotas to hit, no chat traffic that flows through anyone's infrastructure but yours.

WHAT YOU GET

• All your sessions in one place — a unified home lists every chat across every server and agent, newest first, like a messenger. Each row shows the session's own title, its server, and its last message; last-activity is tracked so the order stays right even offline.
• Full-text search across every session you've opened — find a conversation by what was said inside it and jump straight to the matching message.
• Open as many parallel chats as you want, across as many servers as you want.
• Each chat is a real CLI process on your server. Resume sessions across days. Files written by the agent appear on your real filesystem.
• Per-server connection: one authenticated SSH transport is shared by every chat on that host. Pull-to-refresh, agent-status probes, and JSONL prefetch are all free until you disconnect.
• Hardware security keys (FIDO2 / CTAP2): any authenticator that speaks USB-HID FIDO or NFC ISO-DEP. Enroll any number of keys per server. Tap once to connect, then ride the same auth for the rest of the session.
• After your first connect, Conch reconnects silently on launch and on network changes — no extra tap until you actually send a message.
• Software keys too: import existing Ed25519, RSA, ECDSA or DSA private keys, or generate a new Ed25519 keypair on device.
• Background prefetch of session histories so opening an old chat paints instantly from cache.
• Markdown rendering, code-block copy, image and file attachment over an SSH cat-stream, and a status indicator that tells you when work is in flight on the server vs. on your phone.
• Approval modes — SAFE / AUTO / YOLO — mapped to each agent's native sandbox / permission flags so you can decide how much trust the AI gets.
• Memory editor for CLAUDE.md / AGENTS.md / GEMINI.md — global and per-project, edited directly on the server.
• Claude-only extras: subagents browser/editor and custom slash-commands sourced from ~/.claude/commands/*.md, surfaced in autocomplete.

WHAT WE DON'T DO

• We don't proxy your prompts. Your messages go straight from the phone to your SSH server.
• We don't store your credentials anywhere off your device. SSH passwords, private keys, and FIDO credential handles are encrypted with the Android Keystore (AES-256, hardware-backed where the device supports it).
• We don't have a hosted backend. The only network connection Conch opens to anyone other than your own servers is to Sentry, for opt-out crash and performance reporting (no chat content, no identifiers).
• We don't show ads.

WHO IT'S FOR

Software engineers, sysadmins, security researchers — anyone who already pays for a Claude / OpenAI / Gemini account and runs the official CLI on their own VPS, dev box, or rented server. If you're trying Conch before installing one of those CLIs, install it on your server first and run "claude /login" (or the equivalent) before opening the app.

PRIVACY AT A GLANCE

• No accounts.
• No chat content collected.
• Crash and performance reporting is opt-out (Settings → Privacy).
• Full privacy policy: https://nikitaeight24family.github.io/sshai-pages/privacy.html

SUPPORT

Bug reports and feature requests: nikita@eight24family.ai