teveus：终端里的免费开源 AI 编程助手，可运行 Claude Code

teveus is a free, open-source AI coding assistant that runs in your terminal. Run Claude Code, or any AI provider with your own key, in one fast, good-looking app.

Website · Docs · Releases

Install

curl -fsSL https://raw.githubusercontent.com/stawan15/teveus/main/install.sh | sh

Works on macOS and Linux. Run the same line again to update.

Want Claude models on your Claude plan? Install Claude Code from Anthropic and sign in to it. teveus runs your installed Claude Code as it is. You sign in through Anthropic's own login, and teveus never sees your Claude credentials. For OpenAI, Gemini, OpenRouter and others you don't need anything else.

Get started

Open a terminal in your project folder.
Run teveus, then connect an AI with /login: your installed Claude Code, or any provider with your own key. Nothing is connected until you choose.
Type what you want in plain words, like "find the bug in the login page and fix it", and press enter.

teveus shows you every change before it happens. Press y to allow it or n to say no.

The first time, a short setup (3 questions) helps you pick how much guidance you want, a colour theme, and how to connect.

What it looks like

Approve changes before they happen	Every action one search away

Welcome screen with starter ideas	Themes preview live

What you can do


💬 Ask in plain words	Explain code, fix bugs, write tests, review changes
✅ Stay in control	See each change as a coloured diff and approve it first
🔌 Use any AI	Claude Code (with your own Claude account), OpenAI, the Anthropic API, Gemini, OpenRouter (hundreds of models, many free), Groq, DeepSeek, xAI, Mistral, or local models (Ollama, LM Studio)
🖼️ Show, don't describe	Drop a screenshot on the terminal or press `ctrl+v`; long pastes fold into `[Pasted text #1 +40 lines]`
🌐 Web, MCP and helpers	On the Direct API engine: reads web pages, searches the web (Brave or Tavily key), uses MCP servers from `.mcp.json`, and sends research to parallel sub-agents. Claude Code brings its own.
🔒 Rules you set once	On the Direct API engine: "always allow `go test`" is saved per project, and `/permissions` lists, adds and removes allow/deny rules
🧩 Make it yours	On the Direct API engine: your slash commands and skills from `.claude/commands` and `.claude/skills`, `/hooks` to run your own commands around tool calls, background commands for dev servers, and notebook editing
🌿 Sessions that don't collide	`/worktree` gives a session its own git worktree; `/sessions` shows what each one has used
🔎 Find it again	`/search text` looks through this conversation and earlier ones
↩️ Undo	`/undo` puts files back the way they were
🕘 Pick up later	`/resume` reopens earlier conversations
💸 Spend less, when you want	Off until you turn them on: `/concise` for short answers, `/lean` to send fewer tool definitions, and `/minimal` (a slider from off to strict) so the model reuses what exists and writes only what the task needs
✂️ Trim the fat	`/trim` reviews your uncommitted changes for code you don't need; `/trim all` checks the whole project
🪙 Long sessions stay cheap	On the Direct API engine, old tool output is swapped for a one-line note (the model can run it again), unchanged files aren't re-sent, and `/subagent` runs research on a cheaper model
🎨 Make it yours	6 themes (including high contrast), 3 guidance levels, reduce motion

Keys to know

You only need the first three. Press ? inside teveus to see the rest.

Key	Does
`enter`	Send your message
`shift+enter`	New line (or `ctrl+j`, or `\` then `enter`)
`ctrl+k`	Find any action (command palette)
`?`	Show all shortcuts
`shift+tab`	Switch mode: ask first → auto-edit → plan only → autopilot
`@`	Attach a file by name
`ctrl+v`	Paste an image from the clipboard
`esc` `esc`	Stop the current task
drag with the mouse	Select text (it's copied when you let go)
`ctrl+c` `ctrl+c`	Quit

Commands

Type / to see the main ones; keep typing to find the rest. The most useful:

Command	Does
`/login`	Sign in to Claude Code, or connect another AI provider
`/model`	Switch model (search by name, or type `free`)
`/effort`	How hard the model thinks: a slider from auto and low up to max
`/undo`	Undo the last change
`/sessions`	Run several conversations at once: switch, start or close one (`alt+1` to `alt+9` jump to one)
`/resume`	Reopen an earlier conversation
`/search`	Find text in this and earlier conversations
`/worktree`	Start a session in its own git worktree
`/hooks`	Commands that run before or after tool calls
`/diff`	See what changed
`/init`	Let the AI write notes about your project so it works better
`/settings`	All preferences in one place
`/mcp`	See MCP servers, and approve a project's `.mcp.json`
`/permissions`	Saved allow/deny rules for tools
`/update`	Check for a new version and install it
`!command`	Run a shell command, e.g. `!npm test`

Connect other AI providers

Run /login inside teveus and pick one:

OpenRouter: log in with your browser, no key to copy. Hundreds of models; type free in /model to find free ones.
OpenAI, Anthropic, Gemini, Groq, DeepSeek, xAI, Mistral: paste an API key. It's checked before saving and stored only on your computer.
Ollama or LM Studio: runs on your own machine; nothing to set up if it's running.

Use it from scripts

teveus -p "your prompt" runs one prompt without the UI and prints the reply; add -json for the cost and token counts. Tools that need approval are refused unless you pass -mode acceptEdits or -mode auto.

Questions

Do I use my Claude plan, or pay per token?

With the Claude Code engine, teveus runs your installed Claude Code, and Claude Code uses whatever you signed in to it with: your Claude plan or an API key. teveus only starts the claude program and shows what it does. It never reads, stores or sends your Claude login, and the first time it asks you to confirm that you want it to run your Claude Code. With Direct API (/engine), you pay the provider per token with your own key.

Is teveus made by Anthropic?

No. teveus is an independent open-source project. It isn't affiliated with, endorsed by, or supported by Anthropic.

How do I select and copy text?

Drag with the mouse; the text is copied when you let go. Double-click selects a word and triple-click a line. ctrl+y copies the whole last reply. Prefer your terminal's own selection? Run /mouse.

Where are my settings and keys?

In ~/.config/teveus/. Keys are in auth.json, readable only by you.

How do I uninstall?

rm ~/.local/bin/teveus
rm -rf ~/.config/teveus   # also removes settings and saved keys

More ways to install

Go: go install github.com/stawan15/teveus@latest
Windows or manual download: grab an archive from Releases
Installer options: TEVEUS_INSTALL_DIR=/usr/local/bin or TEVEUS_VERSION=v0.1.1 before sh

Development

go build -o teveus . && ./teveus   # run from source
go test ./...                       # tests
docs/record.sh                      # re-record the demo GIF from a real session

Built with Bubble Tea, Lip Gloss and Glamour. CLAUDE.md explains how the code is organised.

Privacy and security

teveus has no telemetry and no servers: what you send goes only to the AI provider you picked. PRIVACY.md lists exactly what is sent where and what is stored on your computer; SECURITY.md explains the safeguards and how to report a vulnerability privately.

What's new

See CHANGELOG.md for every version's changes.

Legal

teveus is an independent project, not affiliated with, endorsed by, or sponsored by Anthropic PBC or any other AI provider. Claude, Claude Code and Anthropic are trademarks of Anthropic PBC; other names belong to their owners and are used only to say what teveus works with.
Your use of Claude Code and Claude is governed by Anthropic's Consumer Terms or Commercial Terms and Usage Policy, and other providers' services by their own terms. teveus doesn't resell, pay for, or route anyone's AI usage; each person uses their own account or key.
AI agents can edit and delete files and run commands. Review what you approve and keep your work in version control. teveus is provided "as is", without warranty of any kind, and the authors aren't liable for any damage, loss or cost from using it; see the license.

License

MIT