English  ·  简体中文

---

A terminal Code Agent purpose-built for DeepSeek prefix caching. Feels like Claude Code — your existing Skills work out of the box. DeepSeek charges up to 120× more for cache misses than hits. Waveloom keeps the System Prompt and message prefix stable at the architecture level, so the longest common prefix stays cache-hot.

curl one-liner (macOS / Linux)

curl -fsSL https://raw.githubusercontent.com/Menfre01/waveloom/main/install.sh | sh

Homebrew

brew trust menfre01/tap
brew install Menfre01/tap/waveloom

Supports macOS / Linux / Windows, AMD64 & ARM64. Installs to ~/.local/bin, no sudo needed.

Windows requires Git for Windows. Open PowerShell and run:

powershell -c "irm https://raw.githubusercontent.com/Menfre01/waveloom/main/install.ps1 | iex"

Then configure your key and start:

waveloom setup
waveloom

Important

API key connects directly to DeepSeek / OpenAI — your code never passes through a third-party server. Every file write and command execution requires your confirmation.

---

How does it compare to Claude Code?

	Waveloom	Claude Code
Cache design	Built for DeepSeek prefix matching: fixed System Prompt, append-only, in-place compaction	Built for Anthropic cache_control: dynamic System Prompt sections, compaction replaces messages
Context compaction	In-place, prefix-stable	Replaces messages with summary
Runtime	Single binary ~18MB	Node.js

Choose Waveloom if: you use DeepSeek, care about API costs, have Claude Code Skills, need a zero-dependency binary
Choose Claude Code if: you use Anthropic, need MCP, are deep in the Claude ecosystem

---

Highlights

• Prefix cache optimized — Fixed System Prompt, append-only message history, four-tier watermark compaction. Maximum common prefix stays cache-hot across turns.
• Permission safety — Three-tier decisions (allow / deny / ask) with pattern-matching rule engine. Every write operation requires your confirmation.
• Session persistence — Close the terminal, come back days later with waveloom --continue. The agent remembers all prior context.
• Plan Mode — Two-stage workflow: explore & design first, implement after approval. Shift+Tab to enter/exit, Guard-enforced write protection, [plan:start]/[plan:end] message pairing for prefix-cache-safe context.
• 9 built-in tools — read_file / write_file / edit_file / shell / web_fetch / ask_user_question / enter_plan_mode / exit_plan_mode / skill — invoked autonomously by the agent.
• i18n multilingual — Full zh-CN / en-US bilingual UI. --locale CLI flag, /locale command, settings.json persistence, auto-detect from LANG env var.
• TUI interactions — @ file references / @ fuzzy file picker / / command palette / /locale switch language / Tab paragraph navigation / Shift+Tab Plan Mode / Ctrl+G theme toggle

---

FAQ

Q: How do I switch models?
Type /model in interactive mode, or waveloom --model deepseek-v4-flash.

Q: Is my API key safe?
Stored locally at ~/.waveloom/. Keys connect directly to DeepSeek / OpenAI — no third-party relay.

Q: How do I switch languages?
Type /locale to toggle between Chinese and English, or waveloom --locale zh-CN. The setting persists automatically in settings.json.

Q: What languages are supported?
Waveloom works with any text-based project. Code verification uses each language's native build tools (go build, npx tsc, cargo build, make, etc.) — no LSP server required.

---

Docs

Document	Content
usage	Interactive mode, shortcuts, Skill system
install	Homebrew / curl / source / shell completions
settings	API key, model, timeout, compaction
prefix-cache	DeepSeek caching, four-tier compaction
environment	Toolchain probing
faq	Frequently asked questions

---

Development

Go 1.25+, make build / make test. See CONTRIBUTING.md for project structure and contribution guide.

---

Apache License 2.0 © 2026 Waveloom Contributors