English | 中文
All-in-One AI-Powered Developer Workstation & Swiss Army Knife for the AI Era
Unifying an Intelligent AI Agent + Audio & Speech Lab + Native Fast Terminal + Local Model Engine + Secret Vault into a single cohesive experience, eliminating the friction of switching across terminals, web browsers, translation tools, and ad-hoc scripts.
- Multi-Model Immersive Chat: Seamlessly interacts with top-tier cloud models and local offline LLMs. Supports real-time Reasoning Effort / chain-of-thought folding, multi-session isolation, syntax highlighting, and instant copy feedback.
- Model Context Protocol (MCP) Ecosystem:
- Full native support for the open MCP standard, unlocking endless external tools and data sources.
- Supports both stdio (local subprocess) and Streamable HTTP / SSE (remote server) transports.
- Fine-grained per-server tool toggles and reusable profile snapshots.
- Built-in Web Search & Deep Fetching:
WebSearch: High-speed concurrent search aggregation across DuckDuckGo, Google, and Bing without browser overhead, extracting titles, URLs, and summaries.WebFetch: Built-in full Chromium rendering environment to scrape dynamic SPAs and JavaScript-heavy pages.
- Human-in-the-Loop Security Approval: Intelligently inspects high-risk commands and file operations (Tool-aware approval), auto-authorizing safe read-only operations while prompting for confirmation on destructive actions.
- All-Scenario Tri-Source Audio Capture:
- 🎤 Microphone Capture: Everyday voice notes and personal speaking.
- 🔊 System Audio Recording: Built on macOS ScreenCaptureKit, with no virtual audio drivers (such as BlackHole or Soundflower) needed to record computer audio directly (video calls, webinars, podcasts, web streams).
- 🔀 Mixed Audio Capture: Records microphone voice + computer speaker playback simultaneously, making it ideal for two-way interviews, meetings, and tutorials.
- Local Offline Whisper Transcription:
- Zero privacy leakage. Dispatches local lightweight and high-precision Whisper models directly on-device with no internet connectivity required.
- Remote / Self-Hosted Streaming STT (R2T2)：
- Sub-second word-by-word streaming transcriptions; built-in repetitive flooding suppression and seamless rollover for extended recording sessions.
- Speaker Diarization & SRT Generation (Nemotron-3 + MMS-Align):
- Deeply integrated with the native high-performance Rust engine
tiginal-diarize, combining acoustic forced alignment with Nemotron-3 speaker separation. - Generates labeled conversational transcripts (
-diar.txt) and standard subtitle files (.srt) with intelligent sentence-level line wrapping to prevent horizontal overflow. - Supports one-click Redo Diarization on existing text and real-time Cancel controls.
- Deeply integrated with the native high-performance Rust engine
- Real-Time & Static Simultaneous Translation (T3PO):
- Bidirectional English-Chinese streaming simultaneous translation with configurable latency modes (
low,native,high). - Supports custom Technical Terms dictionaries (
source=target) and customized Translation Instructions.
- Bidirectional English-Chinese streaming simultaneous translation with configurable latency modes (
- Intelligent Audio Artifact Management:
- Auto-Artifact Discovery: Loading any local audio file immediately scans and loads existing
.txt,-diar.txt, and.srtfiles into their respective tabs. - Granular Artifact Deletion: Delete individual tab files with clear confirmation dialogs displaying the exact filename; multi-file deletion dialogs present each related file on its own line; external audio files hide full-delete buttons to prevent accidental loss.
- Global 3-Second Copy Feedback: Instant green checkmark indicator that locks for 3 seconds before restoring to prevent accidental duplicate clicks.
- Auto-Artifact Discovery: Loading any local audio file immediately scans and loads existing
- Native PTY Performance: Powered by Electron + xterm.js + node-pty, delivering lightning-fast, native command-line responsiveness.
- Split Panes Matrix: Split workspaces horizontally and vertically, navigating between panes with rapid keyboard shortcuts.
- Smart History & Frequency-Based Suggestions: Contextual auto-suggestions sorted by real-world usage patterns.
- Regex Blacklist Filtering: Define custom regex rules to exclude passwords, API tokens, or one-off commands from cluttering shell history.
- Automated Low-Frequency Cleanup: Built-in score-based decay engine that purges stale and rare commands automatically.
- Elegant Themes: Crafted with Catppuccin color palettes and polished micro-animations.
- Smart Model Asset Scanning: Automatically detects local GGUF models, Hugging Face caches, and custom directories with commit-hash revision detection and multi-quantization categorization (e.g. Q4_K_M, Q8_0).
- Resumable Chunked Downloads: Robust Range & ETag-backed downloading engine for multi-gigabyte models that resumes cleanly after network interruptions.
- Engine Lifecycle Supervisor: Directly hosts and schedules local runtimes (such as llama.cpp and tiginal-diarize) while tracking process health and hardware utilization.
- Master Key-Derived Encryption: Uses master-key AES encryption to securely protect all API tokens, SSH keys, and sensitive environment configs.
- Zero-Trust Ephemeral Materialization: Securely injects temporary environment variables and files on demand only during task execution, immediately purging them upon completion to eliminate the risk of committing plaintext
.envfiles.
Tiginal provides extensive customization options through its settings interface:
| Settings Tab | Capabilities & Configuration Possibilities |
|---|---|
| AI Providers | • Pre-configured integrations with OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Groq, OpenRouter, and more. • Native GitHub Copilot authentication workflow. • Connect any OpenAI-compatible local or private endpoints (e.g. vLLM, Ollama, LocalAI). • Automatically syncs models.dev metadata to configure context limits, token limits, and Reasoning Effort. |
| Model Library & Engines | • Scan and index local model directories. • Download GGUF models with automatic resume and version grouping. • Manage local inference runtimes: executable paths, startup flags, and platform optimizations. |
| Speech Providers | • Select STT backends: Local offline Whisper or remote R2T2 streaming endpoints. • Configure preferred languages, concurrency, and rollover policies. • Configure T3PO translation endpoints (WebSocket streaming or LLM backends). |
| MCP & Agent Skills | • Add, edit, and orchestrate MCP servers (local stdio commands or remote SSE URLs). • Enable or disable individual tools with profile snapshots. • Import and mount domain-specific Agent Skills. |
| Credentials Vault | • Master Key initialization, locking, and status monitoring. • Manage isolated environment variable groups (Production, Staging, Local). • Audit materialization events and credential usage logs. |
| Terminal & Shortcuts | • Choose preferred shell binary (zsh, bash, fish, PowerShell). • Customize fonts, font sizes, cursors, and Catppuccin themes. • Configure history scoring parameters and blacklist regexes. • Remap global keyboard shortcuts. |
| System & Permissions | • Diagnose and manage macOS Screen and System Audio capture permissions with guided recovery. |
| Shortcut | Action |
|---|---|
Cmd + T |
New Terminal Tab |
Cmd + W |
Close Tab |
Cmd + 1-9 |
Switch to Tab |
Cmd + \ |
Split Pane Right |
Cmd + Opt + Arrows |
Navigate Across Panes |
Cmd + Shift + W |
Close Current Pane |
Download official prebuilt binaries from GitHub Releases:
- Apple Silicon (M1/M2/M3/M4):
Tiginal-x.x.x-arm64.dmg - Intel Macs:
Tiginal-x.x.x.dmg
macOS may gate the first launch if the build is not notarized or uses ad-hoc signing. If prompted, go to System Settings > Privacy & Security and click Open Anyway. If macOS reports the bundle is damaged, run:
xattr -dr com.apple.quarantine /Applications/Tiginal.appImportant: System Audio Capture Permission
When first attempting system audio recording, check and enable Tiginal under System Settings > Privacy & Security > Screen & System Audio Recording. Quit and reopen Tiginal after modifying permissions.
Prerequisites: Node.js >= 22, Git.
# 1. Clone the repository
git clone https://github.com/ZPVIP/tiginal.git
cd tiginal
# 2. Install dependencies and run tests
npm ci
npm test
# 3. Start development mode
npm start
# 4. Package distribution binaries
npm run distOutput packages will be generated inside the release/ directory.
Use the same Apple signing identity for every build so Accessibility, Microphone, and Screen & System Audio Recording permissions remain valid after an update.
# List available signing identities
security find-identity -v -p codesigningIf the Mac has multiple identities, copy the 40-character SHA-1 value for the identity you want. The SHA-1 value selects one identity without relying on its display name.
Create electron-builder.env in the project root to keep the selection for future builds:
APPLE_SIGNING_IDENTITY=0123456789ABCDEF0123456789ABCDEF01234567Electron Builder loads this file automatically when you run:
npm run distThe repository ignores electron-builder.env because the selected identity is specific to one Mac. To use another identity for one build, set the variable in the command:
APPLE_SIGNING_IDENTITY="FEDCBA9876543210FEDCBA9876543210FEDCBA98" npm run distAPPLE_SIGNING_IDENTITY has priority over CSC_NAME and automatic identity detection. A value from the shell takes priority over the same value in electron-builder.env. If neither variable is set, Tiginal selects a Developer ID or Apple Development identity from the keychain. It uses ad-hoc signing only when no valid identity exists.
Verify the identity after the build:
codesign -dvvv -r- release/mac-arm64/Tiginal.app
codesign -dvvv -r- release/mac/Tiginal.appKeep the selected identity and bundle ID com.tiginal.app unchanged between builds. If you select another identity, macOS requires permission approval once for that identity. If macOS keeps an obsolete permission record, reset it before installing the new build:
tccutil reset Accessibility com.tiginal.app
tccutil reset ScreenCapture com.tiginal.app
tccutil reset Microphone com.tiginal.appDevelopment mode permissions belong to the terminal or IDE that launches Tiginal. Packaged app permissions belong to Tiginal.app.
To cleanly remove Tiginal along with system permissions and caches:
# 1. Reset Screen & System Audio Recording TCC permissions
tccutil reset ScreenCapture com.tiginal.app
# 2. Remove application bundle
rm -rf /Applications/Tiginal.app
# 3. Clean up app configurations, databases, and caches
rm -rf ~/Library/Application\ Support/Tiginal
rm -rf ~/Library/Caches/Tiginal
rm -rf ~/.cache/tiginaltiginal/
├── src/
│ ├── main/ # Electron main process
│ │ ├── audio/ # Audio capture, R2T2, T3PO, AlignmentEngine, PlatformCapture
│ │ ├── models/ # Model library scanner, resumable downloader, engine supervisor
│ │ ├── services/ # MCP client, credential encryption & materialization
│ │ ├── pty.ts # Native PTY orchestration
│ │ └── preload.ts # Secure context bridge
│ ├── renderer/ # React renderer process
│ │ ├── components/Audio/ # Audio workspace, waveforms, transcript/speakers/SRT editors
│ │ ├── components/Chat/ # AI chat interface & MessageBubble
│ │ ├── components/Settings/ # Settings management panels
│ │ └── components/Terminal/ # xterm.js terminal integration & pane management
│ └── shared/ # Shared type definitions and IPC schemas
└── tiginal-diarize/ # High-performance Rust diarization and alignment engine
This project is licensed under the GNU General Public License v3.0 or later (GPL-3.0-or-later) - see the LICENSE file for details.
Third-party libraries, dependencies, and AI model weight licensing details are documented in THIRD-PARTY-NOTICES.md.