ModelSwap — the key & model console for AI coding agents. Vault, provider switching, usage monitoring.
Keys and models, one console. ModelSwap is a local-first open-source tool that manages the full key lifecycle for AI coding CLIs: create → store → switch → verify → sync. All data stays on your machine — no account, no subscription.
| Home · Agent config | Model platforms |
|---|---|
 |  |
| Key vault | Auto-create keys |
|---|---|
 |  |
/model right inside the CLI, no round-trip to ModelSwap.modelswap vault inject --keys ….Ten built-in adapters, each writing that agent's native config files (e.g. config.toml + auth.json for Codex, settings.json for Claude Code):
Claude Code · ChatGPT Codex · OpenCode · OpenClaw · WorkBuddy · ZCode · Hermes · Kimi Code · Grok · MiMo Code
Prefer a GUI? Download the desktop app directly — Apple silicon ｜ Intel — arm64 / x64 dmg installers with sha256 checksums, in-app auto-update, and the bundled browser extension for auto-creating keys. The desktop app and the CLI share the same data directory (~/.modelswap) — pick either, or run both.
Based on both projects' official READMEs as of 2026-09 (cc-switch · codex-router). Different tools, different trade-offs — pick what fits.
| Capability | ModelSwap | cc-switch | codex-router |
|---|---|---|---|
| Positioning | Key & model console (write & exit) | Provider-switch GUI (tray + optional local proxy) | Local model router (background service + control center) |
| Config writing | Field-level merge + pre-switch snapshots, one-click rollback | Atomic writes + auto backups (last 10) + shared snippets | Managed-block injection |
| Request path | Never through ModelSwap | Direct, or via optional proxy (hot-switch/failover) | Routed through the local router service |
| Agents supported | 10 | 8 | Codex-first (Harness/Gemini CLI bridges experimental) |
| Key storage | AES-256-GCM encrypted vault | Local SQLite store (README does not mention encryption) | Stored locally (README does not mention encryption) |
| Auto-create API keys | 31 platforms (browser extension) | — | — |
| Usage queries | 37 subscription/balance sources, direct | Usage dashboard (spend/requests/tokens) | — |
| Multi-device sync | LAN peer-to-peer + 9 self-hosted cloud backends + sync codes | — | — |
| Platforms | macOS / Linux / Windows | macOS / Linux / Windows | macOS / Linux / Windows |
# via npm
npm install -g modelswap
# launch the web console
modelswap web # opens http://localhost:3780
# or from source
git clone https://github.com/Cing-self/modelswap.git
cd modelswap
npm ci --ignore-scripts
npm run build
node dist/main.js web
Everyday commands:
modelswap web # web console (:3780)
modelswap vault set <key> # store a secret interactively (AES-256-GCM)
printf '%s' "$SECRET" | modelswap vault set <key> --stdin # keep secrets out of argv in automation
modelswap vault inject # print export statements (pair with eval)
modelswap provider list # list 42 preset model platforms
modelswap provider switch # interactive provider/model switch per agent
modelswap provider use <provider> # non-interactive switch (script/agent friendly)
modelswap sync pair --create # LAN pairing, or sync push/pull via self-hosted cloud
Shell config boundary: ModelSwap never touches your shell config (
~/.zshrc/~/.bashrc) — no feature writes to it.
The package ships a modelswap agent skill. modelswap skill install /path/to/project installs it into the project's .agents/skills/modelswap/; modelswap skill path prints the built-in source location. The skill documents the parseable read-only commands, non-interactive model switching, and the security boundaries around plaintext keys and cloud sync.
Or install it straight from the public repo via skills.sh:
npx skills add Cing-self/modelswap --skill modelswap
npm ci # install from the lockfile
npm run build # tsc + preset generation + web copy + frontend build
npx vitest run # tests (500+ cases)
cd src/web/frontend && npm run dev # frontend dev server (:5173 → proxies :3780)
Node.js 20+. Frontend: React + TypeScript + Vite; backend: Node (web layer in CommonJS); tests: vitest. See CONTRIBUTING.md for commit and release conventions.
AES-256-GCM encrypted storage, masked display, on-demand terminal injection (vault inject --keys). The first-run wizard scans agent config files and safely imports stray plaintext keys in one click.
42 platform presets (official / aggregator / China-based), 10 agent adapters, multi-endpoint protocols (anthropic / OpenAI-compatible / responses / typesafe), auth-state checks, and three credential modes (subscription / API / third-party). Adding a site starts from an empty model list — you write exactly what you choose.
The browser extension fills and submits key-creation forms inside official consoles (31 platforms). Google AI Studio and Cloudflare channels are temporarily not offered; keys for them can still be added manually.
37 subscription/balance sources queried directly, with threshold alerts (local notifications).
Every agent-config switch is snapshotted automatically; Settings diffs any two snapshots side by side and rolls back in one click.
Does ModelSwap sit on my request path? No. Zero daemons, zero interception: ModelSwap writes config and exits. Your agents talk to model platforms directly — no proxy, no forwarding.
Will switching break my existing settings? No. Surgical writes touch only ModelSwap-owned fields; hooks / statusLine / MCP config stay intact, and every switch is snapshotted for rollback.
Where are my keys stored? Are they safe? AES-256-GCM encrypted locally, with machine-bound key derivation. Nothing ever lands on disk in plaintext (keys found by the import wizard are masked, too). Uninstalling wipes everything.
Where does sync send my keys? LAN mode is peer-to-peer. Cloud sync goes to storage you create yourself (Cloudflare / Supabase / WebDAV…) and payloads are encrypted — the provider never sees plaintext. You can leave sync entirely off.
Which agents are supported? 10 adapters: Claude Code, Codex, OpenCode, ZCode, Kimi Code, Grok, Hermes, OpenClaw, Mimo Code, WorkBuddy.
How do I uninstall?
npm uninstall -g modelswap, then remove ~/.modelswap. Agent configs already written keep working unchanged.
ModelSwap is released under the MIT License. © Cing-self / ModelSwap contributors (2026).
FAQs
ModelSwap — the key & model console for AI coding agents. Vault, provider switching, usage monitoring.
The npm package modelswap receives a total of 61 weekly downloads. As such, modelswap popularity was classified as not popular.
We found that modelswap demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Research
/Security News
Socket uncovered two malicious VS Code themes in a GlassWorm-linked cluster with thousands of installs across VS Code Marketplace and Open VSX.
Security News
/Company News
Capital One is partnering with Socket to proactively secure its open source supply chain.
Security News
Socket CTO Ahmad Nassri discusses how to keep AI agents from bypassing package blocks, limit credential access, and monitor their actions.