Providers

A provider is what answers. Connect as many as you like: Conch drives all of them at once, from one model picker.

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

Every provider at once

There is no "current provider" to switch. Each one you connect adds its models to the same picker, grouped by provider and searchable by name.

One is the default. New chats start there. Change it with Make default in Settings → Providers, or from the picker.
A chat can change provider mid-way. The one that joins is handed what it missed, word for word, and what was done along the way: which tools ran and how each went, the browser’s page, files changed, an open plan. So it carries on rather than starts again.
Each provider carries its own chat on. Claude Code, Codex, Codex CLI and the programs that can (Copilot, Gemini CLI, Grok) pick a chat up where they left it. If one can’t, it’s given the whole conversation instead.
What you set up belongs to Conch. Your apps, skills, memory and routines work with every provider. What a provider brings by itself is shown apart, and says so.

What each one can do

Providers differ. Conch never pretends otherwise: a feature a provider can't do is hidden or explained, never broken.

Provider	Your files and commands	Saves memories itself	Asks before each step	Works offline
Claude Code	Yes	Yes	Yes	No
Codex CLI	Yes	Yes	Yes	No
Codex	Yes	Yes	Yes	No
GitHub Copilot	Yes	Yes	Yes	No
Gemini CLI	Yes	Yes	Yes	No
Grok	Yes	Yes	Yes	No
On this computer	No	Yes	Yes	Yes
LM Studio	No	Yes	Yes	Yes
OpenRouter	No	Yes	Yes	No
Anthropic API	No	Yes	Yes	No
OpenAI	No	Yes	Yes	No
Google Gemini	No	Yes	Yes	No
xAI API	No	Yes	Yes	No
DeepSeek	No	Yes	Yes	No
Mistral	No	Yes	Yes	No
Groq	No	Yes	Yes	No
Cerebras	No	Yes	Yes	No
Z.ai	No	Yes	Yes	No
Kimi	No	Yes	Yes	No
MiniMax	No	Yes	Yes	No
Qwen	No	Yes	Yes	No
Ollama Cloud	No	Yes	Yes	No

Connecting one

Open Settings → Providers. Yours are on top. The rest wait below as tiles, sorted by what connecting takes, and you can find one by name or by what it's good at.

Coding agents. Claude Code and Codex CLI bring their own tools: they run commands and change files in your folders, on a plan you already pay for. Every command and change asks through Conch first, as your chat's mode says.
Your plans. Codex, GitHub Copilot, Gemini CLI and Grok use a plan you already pay for, through the provider's own program on this computer, with Conch's tools doing the work. If it's missing, Install gets it, with the command shown; then you sign in with the program's own sign-in. Conch never sees its credentials.

Your Claude Pro or Max plan works in Conch through Claude Code, Anthropic's own program. Anthropic's terms keep those sign-ins to Claude Code and Claude.ai, so Conch offers no other way to use them. The Anthropic API card is pay as you go, with a Console key.

On this computer. A model on this computer, through Ollama or LM Studio, or a server you run yourself.
Pay as you go. Paste a key and Conch checks it before keeping it. "Saved" means it works.

Each provider opens on a page of its own, and stays there once it connects so you can see it worked. Providers, at the top of that page or in Settings' list, goes back to them all. Every place in Settings has its own address, like /settings/providers, so reloading, a bookmark or the browser's back button lands where you were.

Have a key? Paste it anywhere

Paste a key anywhere on Settings → Providers. When it starts with something only one company uses (gsk_ is Groq, xai- is xAI), Conch knows whose it is and checks it with that provider straight away. Plenty of keys just start sk-, and some start with nothing at all; for those, Conch asks whose it is first. A key is only ever sent to the company it belongs to, never tried at several to see which one takes it.

A company with regions (Kimi, Z.ai, MiniMax, Qwen) is tried at each of its own addresses, and Conch remembers the one that took your key.

Found on this computer

Conch looks for what's already here: a provider's key in this computer's settings (OPENAI_API_KEY), or a model server running on its usual port. Each is offered under Found on this computer, ready in one press. Nothing is used until you press it, and a key is never shown.

Where keys live

A key stays on this computer, in a file only you can read, and is never shown again. Or keep it in 1Password: paste a reference (op://Vault/Item/field) instead of the key, and Conch asks 1Password for it only when a chat needs it.

When one can't answer

Offline, or at a usage limit, Conch hands the turn to the provider you chose for that, and says so in one line. See Offline and at a limit.