KeyPilot

Local credential vault & quota tracker for the AI era.
Windows-first · Rust + Tauri 2 · MIT

简体中文 · License

---

Note

This repository is a public showcase snapshot of KeyPilot V0.2.1. Issues and pull requests are welcome.

Contents

• Quick Start
• About
• Features
• Connecting to Providers
• Tech Stack
• Scope (V0.2.1)
• Project Structure
• Development
• When (Not) to Use
• Roadmap
• License

Quick Start

Prerequisites

Tool	Version
Rust (nightly)	1.92
Node.js	22.22
pnpm	11.0.9
Tauri CLI	2.11.0
WebView2	bundled with Windows 11
MSVC Build Tools	required for cargo linker

Run from source

git clone https://github.com/YHlorra/keypilot.git
cd keypilot

# 1. install frontend deps
cd webui && pnpm install && cd ..

# 2. start dev (Tauri hot-reload, opens native window)
cargo tauri dev

The first launch seeds an empty SQLite DB at %APPDATA%\com.keypilot.app\. Add a provider, paste an API key, hit Fetch Quota — done.

About

KeyPilot is a local credential manager for AI power users. It centralises API keys, AK/SK pairs and connection strings for every LLM provider you touch, and refreshes quota / balance in the background so you know what's left before you hit send.

V0.1 deliberately stops at three things:

1. Provider presets — every LLM / Coding-Plan provider you actually use has a built-in field schema.
2. Live quota — the things you pay for (API wallet, Coding Plan 5-hour / weekly windows) update in the background and surface in a single tray-side panel.
3. Token usage history — auto-parses sessions from OpenCode / Claude Code / Codex, so you can see which agent burned what, when.

Credential encryption, cross-platform support and cloud sync are explicitly out of scope — see Roadmap and When (Not) to Use.

Features

• 21 built-in provider presets (V0.2.1) — Anthropic, OpenAI, DeepSeek, GitHub Models, Volcengine, Kimi, GLM, MiniMax (CN/EN), Mimo, OpenRouter, Groq, Mistral, SiliconFlow, Together, StepFun, Cohere, Perplexity, Fireworks AI, AI21 Labs. Each preset ships its own field schema, docs_url and quota endpoint.
• Multi-endpoint catalog (V0.2.1) — 6 dual-protocol presets (Kimi / GLM / Volcengine / DeepSeek / MiniMax-CN / MiniMax-EN) expose both OpenAI-compat and Anthropic-compat endpoints under a single API key. Click "另有 N 个协议端点" on the card to inspect.
• Add Custom Provider (V0.2) — + → "+ 自定义" dialog (shadcn/Radix/Tailwind). Pick protocol, fill URL + Key, "测试连接" runs provider.preflight against the live endpoint before save.
• 7 Coding Plan trackers — Kimi For Coding, GLM Coding, MiniMax Token Plan (CN+EN), Volcengine Ark Coding Plan, ZenMux 5h + weekly windows. Both 5-hour and weekly windows surfaced side-by-side.
• 5 protocol registry (V0.2) — typed match on ProtocolId (OpenAI / Anthropic / GitHub / Balance / DeepSeek). Adding a new protocol = 1 file under src-tauri/src/provider/protocols/ + 1 line in registry.rs. Adding a new preset = 1 toml file. No central switch.
• 3 themes — Dark / Light / Follow System. Radix UI Colors palette, brutalist typography, no Tailwind defaults.
• Visibility tri-state — visible → masked → revealed. The disk never cares; the UI decides.
• Token usage history — auto-parses OpenCode / Claude Code / Codex session logs. Heatmap + trend line + per-agent leaderboard.
• System tray resident — single source of truth, 5-minute staleTime, real-time event push from Rust backend.
• Action registry — extension hooks for power users (custom fetchers, auto-refresh, low-quota alerts).

Connecting to Providers

Every provider follows the same flow: pick a preset → fill in api_key (and any preset-specific fields) → Fetch Quota.

Category	Examples	What you need
LLM API	Anthropic, OpenAI, DeepSeek, Kimi, GLM, MiniMax	api_key
LLM Aggregator	OpenRouter, ZenMux, Volcengine Ark	api_key (+ optional endpoint)
Coding Plan	Kimi For Coding, GLM Coding, MiniMax Token Plan, Volcengine Coding Plan, ZenMux	api_key (auto-discovers 5h / weekly windows)
Dev / Data	GitHub Models, Postgres	api_key / connection string

Tip

Provider presets live in src-tauri/src/provider/. Adding a new provider = one file (adapter) + one row in the preset registry. No central switch statement to touch.

For Coding Plans, KeyPilot auto-fetches both windows and shows the tighter one in red when it drops below 20%.

Tech Stack

Layer	Tech
Desktop shell	Tauri 2
Backend	Rust (edition 2021, nightly 1.92)
Frontend	React 18 + TypeScript
Build	Vite 6
State	TanStack Query v5
UI	shadcn/ui (Radix Primitives) + Tailwind + @radix-ui/colors
DB	SQLite (rusqlite bundled)
Tray / FS watch	tauri::tray, notify-debouncer-full

Scope (V0.2.1)

Capability	Status
Windows 10 / 11	Supported
Credential storage	Plaintext + Windows ACL (no encryption)
Live quota	21 LLM presets + 7 Coding Plans (V0.2.1)
Multi-endpoint per preset	V0.2.1 — 6 dual-protocol presets share 1 API key
Token usage history	OpenCode / Claude Code / Codex auto-parse
Cross-platform	Not in V0.3+
Master password / Argon2	Deferred to V0.2.2+ RFC
Auto-refresh / low-quota alerts	V0.2
Real LLM call routing	V0.3+ (catalog already has extras plumbing)
Import / Export	V0.2

Project Structure

keypilot/
├── README.md / README.zh-CN.md   # this file
├── LICENSE                        # MIT
├── src-tauri/                     # Rust backend (Tauri 2)
│   ├── Cargo.toml / tauri.conf.json / build.rs
│   └── src/
│       ├── main.rs / lib.rs / database.rs / store.rs / error.rs / types.rs
│       ├── provider/              # 24 provider adapters
│       ├── services/              # provider / category / quota / token_usage / pricing / auto_import
│       ├── commands/              # tauri IPC commands
│       ├── actions/               # Action Registry (extension hooks)
│       └── tray.rs                # system tray
└── webui/                         # React 18 + TS + Vite 6
    ├── package.json / vite.config.ts
    └── src/
        ├── components/ (30+ files)
        ├── pages/ (UsagePage)
        ├── hooks/ (useUsage / useProviders / useQuota / ...)
        ├── lib/ (api / utils / action-registry / format)
        └── styles/ (globals.css)  # design tokens

Development

# dev (hot reload)
cargo tauri dev                  # run from project root

# frontend only
cd webui && pnpm dev

# typecheck
cd webui && pnpm tsc --noEmit

# unit tests (Rust)
cargo test --manifest-path src-tauri/Cargo.toml

# unit tests (WebUI)
cd webui && pnpm test

# e2e (Playwright, when configured)
cd webui && pnpm exec playwright test

# build release
cargo tauri build

When (Not) to Use

Use KeyPilot when:

• You juggle 3+ AI providers / Coding Plans and want one place to see quota left.
• You want to track how much OpenCode / Claude Code / Codex sessions cost you per agent per day.
• You run Windows 10/11 and trust the local-user ACL boundary.

Do not use KeyPilot when:

• You need cross-platform (macOS / Linux) — V0.3+ at the earliest.
• You need at-rest encryption (master password, Argon2id, DPAPI, SQLCipher) — V0.1 explicitly skips this.
• You need a general-purpose password manager (1Password / Bitwarden / KeePass are the right tools).
• You need cloud sync, multi-device, or team sharing — out of scope.
• You run on a shared / low-trust Windows host where anyone can read %APPDATA% of your user.

Roadmap

Version	Status	Focus
V0.1	✅ Shipped	5 presets, manual quota, Windows-only, plaintext credentials
V0.2	✅ Shipped	18 presets via catalog; 5 protocol registry (typed match); AddCustomProviderDialog + provider.preflight
V0.2.1	✅ Shipped	Multi-endpoint catalog — 6 dual-protocol presets (Kimi / GLM / Volcengine / DeepSeek / MiniMax-CN / MiniMax-EN) merge 12 toml → 6 with [[extras]]; +3 single-protocol presets (Mimo / Fireworks / AI21) → 21 total; ProviderCard extras foldout; preset icons auto-resolved from catalog
V0.2.2	Planned	Encryption RFC (SQLCipher / master-password Argon2id / DPAPI); T3.9 advanced JSON mode
V0.3	Planned	macOS / Linux support; real LLM call routing using catalog extras
V1.0	Planned	Stable API, signed releases, auto-update channel

License

MIT © 2026 KeyPilot Authors.