Skip to the page

DocumentationDevelopment · be97819e

Everything Conch does, in plain words.

A guide for each thing it does, and a reference read straight from the code, so it is never behind it.

curl -fsSL https://conchagent.com/docs/next/install.sh | CONCH_BRANCH=main sh

Get started How Conch works

Find your way

Eight short sections. Start at the top, or jump in.

Get startedInstall it, say hello, take it with you.ProvidersThe assistants and models Conch drives.Talk to me hereReach it from the chat apps on your phone.What it doesMemory, skills, the web, your files.It looks after itselfRepair, backups, updates, undo.SecurityWho gets in, and what the assistant may do.ReferenceEvery command, setting and message.Under the hoodHow Conch is built, and why.

Every provider, at once

One model picker for all of them, and a chat can move between them without losing its thread.

Coding agents. They bring their own tools and work in your folders, on your plan. They ask through Conch.

Claude CodeYour Claude plan, on this computer Codex CLICodex, with its own tools

Your plans. A plan you already pay for, signed in with its own program.

CodexUse your ChatGPT plan GitHub CopilotEarly supportUse your Copilot plan Gemini CLIEarly supportUse your Google account GrokEarly supportUse your Grok plan

On this computer. Private, free, and answers with no internet.

On this computerFreePrivate, free, and works offline LM StudioFreeYour LM Studio models, here Another serverAny OpenAI-compatible address

Pay as you go. A key you paste. Conch checks it with that company before keeping it.

OpenRouterFree modelsHundreds of models, one key Anthropic APIClaude, pay as you go OpenAIGPT, billed per use Google GeminiFree tierGemini, free to start xAI APIGrok, billed per use DeepSeekStrong models, tiny prices MistralFree to startEuropean models, free to start GroqFree tierOpen models, instantly fast CerebrasVery fast open models Z.aiGLM models, very cheap KimiKimi models, very long context MiniMaxStrong agents, huge context QwenFree starter quotaQwen models, tiny to frontier Ollama CloudFree planBig open models, hosted

In the apps you already use

A bot of your own, no public address, and nobody gets in unless you let them.

TelegramAbout 2 minThe easiest. Make a bot, paste its key, say hello.DiscordAbout 4 minMessage your assistant privately, from any device.SlackAbout 4 minA private DM with your assistant, in your workspace.WhatsAppAbout 1 minYour own WhatsApp. Scan a code, then message yourself.SignalAbout 2 minYour own Signal. Scan a code, then use Note to Self.iMessageAbout 1 minText yourself from your iPhone. Nothing to make.EmailAbout 3 minWrite to yourself+conch, from any mail app.Microsoft TeamsAbout 8 minA private chat with your assistant, at work.MatrixAbout 3 minEncrypted, on the homeserver you choose.WeChatAbout 5 minThrough a WeCom bot, or your own Official Account. 微信 SMSAbout 5 minA phone number of its own. Text it from any phone.MattermostAbout 3 minA bot on your own Mattermost server. No public address.LINEAbout 6 minA LINE Official Account of your own, answering you.Rocket.ChatAbout 3 minA bot on your own Rocket.Chat server. No public address.Google ChatAbout 12 minA Chat app of your own, in your Google Workspace.