VibeCoding

Juejin's Local AI Gateway v1.1.0 Consolidates Model Routing, Usage Billing, and Tool Integrations

By XCaptain · Nov 12, 2026

Read original on juejin.cn ↗ Google Translate ↗ Alt translation

Managing multiple AI subscriptions and tool chains from one local panel removes the overhead of switching accounts and reconciling bills, and keeps credentials and usage data on the user's own machine rather than in a cloud dashboard.

Summary

Juejin's desktop AI gateway v1.1.0 consolidates provider credentials, model routing, usage history, and local runtime tools into one panel. It adds a standalone gateway window and tray usage page, lets you search and switch default models, and supports custom API providers, LAN sharing, and port changes. Model connectivity tests run in one click, and the catalog now uses short aliases while keeping old long IDs compatible.

The update hardens security by moving the encryption master key to a local file at ~/.ai-usage/agent-router/vault.key, so the system keychain isn't required for daily work. Protocol coverage now includes Messages, Responses, Chat, and Gemini, with a built-in search execution layer and cross-protocol tool calls. Direct integration for DeepSeek Harness is added, and runtime tools detect and reversibly configure Claude Code, Codex, ZCode, OpenCode, MiniMax Code, Pi, and Claude Desktop.

Usage tracking gets new collectors for CodeArts Agent and PI-Desktop, and Trae international accounts read quotas from an existing IDE login without extra session keys. A Grok Bot subscription card appears without needing Cursor, Qoder pools show remaining percentages, and sync status distinguishes local vs. cloud states. Pricing from models.dev is synced for newer models, and fixes address DSH Desktop discovery, cursor.json bloat, and macOS installer issues.

Takeaways

— The encryption master key is stored at ~/.ai-usage/agent-router/vault.key instead of the system keychain.

— Native protocol support spans Messages, Responses, Chat, and Gemini, with a built-in search execution layer for cross-protocol tool calls.

— DeepSeek Harness connects directly, automatically writing profile and credentials and allowing rollback.

— Runtime tool integration covers Claude Code, Codex, ZCode, OpenCode, MiniMax Code, Pi, and Claude Desktop with reversible configuration.

— New usage collectors cover CodeArts Agent and PI-Desktop.

— Trae international account quotas and usage read from an existing IDE login, no extra session key needed.

— A Grok Bot subscription quota card works without installing Cursor, and Qoder pools display remaining percentages.

— Sync status now distinguishes local collection, cloud upload, pending, failure, and unreported states.

— models.dev official pricing is synced for models including Claude Fable, Gemini 4 Carbon, Muse Ember, GPT-6.1 Terra, and Mistral Large 4.

— Fixes address DSH Desktop custom data directory discovery, limited Codex/ZCode snapshot retention to stop cursors.json growth, and a macOS installer intermediate image issue.

Conclusions

Storing the master key in a local file rather than the OS keychain trades system-level integration for portability across dev and release builds, a deliberate move to keep the gateway working without OS prompts.

The gateway's roster of supported coding agents (Claude Code, Codex, OpenCode, and others) positions it as a single control plane for AI development tools, not merely a usage tracker.

Concepts & terms

Local AI Gateway

A desktop application that centralizes configuration, routing, and usage tracking for multiple AI model providers, managing credentials and tool integrations locally.

Model routing

The process of directing API requests to different AI models based on rules, availability, or user preference, handled centrally by the gateway.

Source: juejin.cn ↗ Google Translate ↗ Backup ↗