APIBypass

One BaseURL. All models. Zero hassle.

APIBypass is a macOS menu bar app that acts as your local LLM API gateway. Configure once, and every AI client connects to all your models through a single address — automatic format translation, model mapping, parameter injection, and Claude Code multi-model launch, all in one place.

English · 简体中文

Why APIBypass?

Tired of these headaches?

Juggling multiple API keys across different clients and providers
Want Claude Code to use third-party models, but it only speaks Anthropic format
Different models need different temperature, thinking, and token settings — constantly tweaking configs
Claude Code has cache hit issues and can't properly recognize long-context models
API keys scattered everywhere, security feels like an afterthought

APIBypass solves all of these. One app, one address, done.

Highlights

One Gateway for All Models

Configure each provider's BaseURL and API key once. Your clients only need http://127.0.0.1:8390 — DeepSeek, Qwen, Kimi, OpenAI, Anthropic, and more. One entry point, no more config juggling.

Automatic Format Translation

Client sends Anthropic format, upstream expects OpenAI? Translated automatically, completely transparent. Requests, responses, SSE streams, tool calls, thinking mode — all handled. Your client doesn't even know a format translation happened.

Custom Models & Fine-Grained Control

Model mapping: your client requests claude-sonnet-4-6, APIBypass calls whatever model you specify
Parameter injection: temperature, max tokens, thinking mode — controlled in one place
Thinking mode toggle: one-click on/off, compatible with both Anthropic and OpenAI formats

Claude Code Multi-Model Launcher

Break Claude Code's model barrier. Launch Claude Code with one click and assign different providers to Opus, Sonnet, Haiku, and other roles — one session, multiple providers working together seamlessly.

Fixes for Claude Code Known Issues

Addresses known compatibility issues when using third-party models with Claude Code, including cache optimization and long-context model recognition — making third-party models more stable and reliable in Claude Code.

Security & Privacy

API keys stored in macOS Keychain, no plaintext in config files. All traffic stays local, no third-party servers, no telemetry.

Features

API Format Translation

Automatic Anthropic ↔ OpenAI format translation — request bodies, responses, SSE streams, tool calls, thinking mode
Smart detection: only translates when client format ≠ upstream provider format
Supports /v1/chat/completions (OpenAI), /v1/messages (Anthropic), /v1/responses (OpenAI Responses API)

Bypass Mode

One-click toggle in the menu bar to enable pure proxy mode
When activated, requests pass through transparently without API format conversion, while still preserving model mapping and parameter injection
Useful when upstream provider natively supports the same format as the client — eliminates unnecessary translation overhead

Claude Code Launcher

One-click launch from menu bar with all environment variables pre-configured
Multi-provider model assignment: assign different providers to Opus, Sonnet, Haiku, and Subagent roles in a single session
Terminal selection: Terminal.app, iTerm2, Alacritty, Kitty, Warp, Hyper
Effort level selector: none, low, medium, high, max
Cache fix: strips cch billing headers, controls CLAUDE_CODE_ATTRIBUTION_HEADER
1M context fix: auto-appends [1m] suffix for long-context models

Provider Management

Separate Provider configurations (API type, base URL, API key)
Model mappings reference providers — no duplicate credentials
Environment variables per provider for Claude Code integration
Auto-migration from legacy format

Core Proxy

Runs in the macOS menu bar — stays out of your Dock
Local proxy server on 127.0.0.1:8390 (configurable)
OpenAI Chat Completions API (/v1/chat/completions)
Anthropic Messages API (/v1/messages)
OpenAI Responses API (/v1/responses)
SSE streaming support — real-time forwarding when stream: true
Claude Code compatibility mode — strips cch billing headers and controls CLAUDE_CODE_ATTRIBUTION_HEADER

Model Mapping

Model name mapping (client request name → actual model)
Multiple configurations, each independently switchable
Enable/disable individual configurations at the top of each config page
Right-click context menu: copy config (including API key), delete config

Parameter Injection

Temperature, Max Tokens, Top P, Frequency Penalty, Presence Penalty
Thinking mode override: toggle on/off for Anthropic (thinking parameter) and OpenAI-compatible APIs (enable_thinking parameter)
Thinking budget control (Anthropic format)
Custom JSON fields — inject any parameter with any value type (supports strings, numbers, booleans, objects, arrays)

Security & Privacy

API Keys stored securely in macOS Keychain — single authorization, no plaintext in config files
All traffic processed locally — no third-party servers involved
No telemetry or usage data collected

UI & UX

Bilingual interface: Chinese (中文) and English, switchable in Settings
Three-column layout: Providers sidebar, detail editor, mapping overview
Formatted JSON request logging in terminal
Settings panel with about info, version number, GitHub link (clickable), and License

System Requirements

macOS 14.0 or later

Installation

Download (Recommended)

Download the latest DMG from the Releases page. Mount the DMG and drag APIBypass to your Applications folder. On first launch, allow network connections when prompted.

Build from Source

Build requirements: Swift 6.0+ / Xcode 16.0+ (or Command Line Tools only)

git clone https://github.com/panando/APIBypass.git
cd APIBypass

# Run in debug mode
swift run

# Or build release binary
swift build -c release
.build/arm64-apple-macosx/release/APIBypass

On first launch, allow network connections when prompted.

Usage

1. Start the Server

Click the APIBypass icon in the menu bar. The server starts automatically on launch, and the indicator turns green when running on 127.0.0.1:8390 (default port, configurable in Settings). You can also manually start/stop via the menu.

2. Configure Providers

Click "Configure..." in the menu bar. Create a provider:

Field	Description	Example
Provider Name	A label for this provider	`My DeepSeek`
API Provider	OpenAI or Anthropic	`OpenAI`
Base URL	Upstream API endpoint	`https://api.deepseek.com/v1`
API Key	Your upstream API key	Stored in Keychain

The API Provider type determines whether format translation is needed:

If client sends Anthropic format but provider is OpenAI → automatic translation
If client sends OpenAI format but provider is Anthropic → automatic translation
Same format → no translation, direct passthrough

3. Configure Model Mappings

Inside each provider, create model mappings:

Field	Description	Example
Config Name	A label for this mapping	`Claude Sonnet`
Incoming Model	The model name your client sends	`claude-sonnet-4-6`
Actual Model	The real model to call upstream	`deepseek-chat`

The master enable switch at the top of each config page controls whether that mapping is active.

4. Parameter Injection

Reasoning Mode Override: Toggle the master switch to control thinking mode:
- Enable → inject enable_thinking: true (OpenAI) or thinking: {type: "enabled"} (Anthropic)
- Disable → inject enable_thinking: false or thinking: {type: "disabled"}
- Turn off → don't touch, use API default
Standard Parameters: Fill in Temperature, Max Tokens, Top P, Frequency Penalty, or Presence Penalty to inject.
Custom Fields: Inject arbitrary JSON key-value pairs. Values are auto-detected: "true"/"false" → boolean, "42" → integer, "3.14" → double, "{\"key\":\"val\"}" → JSON object.

5. Launch Claude Code

Click "Launch Claude Code" in the menu bar:

Select a provider (base URL and API token are auto-configured)
Select your preferred terminal
Choose a model for ANTHROPIC_MODEL (required)
Optionally configure Opus/Sonnet/Haiku/Subagent model defaults
Set effort level (none, low, medium, high, max)
Optionally set a working directory
Click "Launch" — Claude Code opens in your terminal with all environment variables set

6. Configure Your Client (Manual)

If not using the launcher, point your AI client to http://127.0.0.1:8390/v1:

Example (Cursor):

OpenAI Base URL: http://127.0.0.1:8390/v1
Anthropic Base URL: http://127.0.0.1:8390/v1

The API Key field can be anything — the proxy replaces it with your real key.

7. Verify (Developers)

When running with swift run, watch the terminal for formatted request logs:

Incoming request body (original)
Upstream URL and actual model name
Transformed request body (with injected parameters)
Format translation status (if applicable)

This step is for developers. Downloaded DMG users do not see terminal output.

Settings

Access via menu bar "Settings...". The settings panel provides:

Language: Switch between Chinese (中文) and English — takes effect immediately
Server Port: Configure the local proxy port (default 8390, restart required to apply)
About: Version number, project description, GitHub repository link (clickable), and MIT License info

Project Structure

APIBypass/
├── APIBypassApp.swift          # App entry point + menu bar icon
├── Core/
│   ├── ConfigManager.swift     # Config management (UserDefaults)
│   ├── FormatTranslator.swift  # Anthropic ↔ OpenAI request/response translation
│   ├── HTTPServer.swift        # Hummingbird HTTP server + SSE streaming
│   ├── LocalizationManager.swift  # i18n: Chinese/English strings
│   ├── ProxyEngine.swift       # Request transform engine
│   └── StreamTranslator.swift  # SSE streaming format translation
├── Models/
│   ├── APIProvider.swift       # API provider enum
│   ├── ProviderConfig.swift    # Provider + environment variables
│   ├── ModelMapping.swift      # Data models
│   └── RectifierModels.swift   # Claude Code compatibility helpers
├── Services/
│   ├── ClaudeCodeLauncher.swift  # Terminal detection + launch
│   ├── KeychainService.swift   # Keychain storage with caching
│   └── NetworkService.swift    # HTTP + streaming network service
├── UI/
│   ├── ConfigWindow.swift      # Config window + new mapping sheet
│   ├── MenuBarView.swift       # Menu bar popup
│   ├── SettingsView.swift      # Settings panel (language + port + about)
│   └── Views/
│       ├── EnvironmentVariablesCard.swift  # Provider env vars config
│       ├── LaunchClaudeCodeView.swift      # Claude Code launcher UI
│       ├── MappingCardView.swift           # Expandable mapping card
│       ├── MappingDetailView.swift         # Config detail editor
│       ├── MappingEditForm.swift           # Shared form component
│       ├── MappingListView.swift           # Config list with context menu
│       ├── NewMappingView.swift            # New mapping sheet
│       ├── NewProviderView.swift           # New provider sheet
│       └── ProviderDetailView.swift        # Provider detail editor
└── Package.swift               # Swift Package manifest

Tech Stack

SwiftUI — macOS menu bar app + windows
Hummingbird 2.0 — HTTP server framework
Keychain Services — API key secure storage with caching
UserDefaults — Config + language persistence
async/await — Async networking (including SSE streaming)
ServiceLifecycle — Service lifecycle management

Privacy

API Keys are stored in the system Keychain and never uploaded anywhere
All traffic is processed locally — no third-party servers involved
No telemetry or usage data collected

Star this project

Your support means a lot — please star this project if you find it useful.

Name		Name	Last commit message	Last commit date
Latest commit History 225 Commits
APIBypass		APIBypass
APIBypassTests		APIBypassTests
docs/superpowers		docs/superpowers
.gitignore		.gitignore
APIBypass.png		APIBypass.png
CLAUDE.md		CLAUDE.md
Info.plist		Info.plist
LICENSE		LICENSE
Package.swift		Package.swift
README.md		README.md
README_CN.md		README_CN.md
RELEASE_NOTES.md		RELEASE_NOTES.md
build-app.sh		build-app.sh
build-dmg.sh		build-dmg.sh
icon.icns		icon.icns
menu.png		menu.png
screenshot_configure.png		screenshot_configure.png
screenshot_launch.png		screenshot_launch.png
screenshot_settings.png		screenshot_settings.png

Folders and files

Latest commit

History

Repository files navigation

APIBypass

Why APIBypass?

Highlights

One Gateway for All Models

Automatic Format Translation

Custom Models & Fine-Grained Control

Claude Code Multi-Model Launcher

Fixes for Claude Code Known Issues

Security & Privacy

Features

API Format Translation

Bypass Mode

Claude Code Launcher

Provider Management

Core Proxy

Model Mapping

Parameter Injection

Security & Privacy

UI & UX

System Requirements

Installation

Download (Recommended)

Build from Source

Usage

1. Start the Server

2. Configure Providers

3. Configure Model Mappings

4. Parameter Injection

5. Launch Claude Code

6. Configure Your Client (Manual)

7. Verify (Developers)

Settings

Project Structure

Tech Stack

Privacy

Star this project

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 26

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages