Kotro Proxy Engine

Transparent IDE sidecar for the Kotro Proxy Engine — a local LLM gateway with streaming semantic cache, PII redaction, and context compression for OpenAI and Anthropic APIs.

Works in VS Code, Cursor, and other VS Code–compatible editors.

Features

Zero-config sidecar — spawns the native proxy binary on startup
Semantic SSE cache — faster repeat prompts; X-Kotro-Cache: HIT on cache hits
Status bar — live cache result and compressor bytes saved (polls every 5s)
Operator dashboard — open from the status bar (http://127.0.0.1:9090/dashboard)
Isolated telemetry — /metrics and /dashboard bind to loopback by default, separate from LLM traffic
Context-aware cache keys — window_n strategy prevents false cache hits in multi-turn agent loops

Screenshots

Status bar

Kotro status bar showing cache hit and bytes saved

Dashboard

Kotro proxy operator dashboard

Install

Install from the Marketplace (Install above), or:

code --install-extension kotrolabs.kotro-proxy-engine

Reload the window. The sidecar starts automatically.
Point your AI client at http://localhost:8080/v1 (OpenAI-compatible base URL).

Quick start (Cursor / VS Code)

Set your provider API key in the environment or your agent config.

Configure the extension (optional):

Setting	Default	Maps to
`kotrolabs.listenAddr`	`:8080`	`KOTRO_LISTEN_ADDR`
`kotrolabs.metricsAddr`	`127.0.0.1:9090`	`KOTRO_METRICS_ADDR`
`kotrolabs.upstreamUrl`	`https://api.openai.com`	`KOTRO_UPSTREAM_URL`
`kotrolabs.enableCache`	`true`	`KOTRO_ENABLE_CACHE`
`kotrolabs.enableRedaction`	`true`	`KOTRO_ENABLE_REDACTION`
`kotrolabs.enableCompression`	`true`	`KOTRO_ENABLE_COMPRESSION`
`kotrolabs.enableMetrics`	`true`	`KOTRO_ENABLE_METRICS`

Click the Kotro item in the status bar to open the dashboard.

Verify it works (2 minutes)

The extension starts the proxy. Your IDE must send API traffic to it.

Step	Action	Success signal
1	Cmd+Shift+P → Kotro: Verify Cache	Notification: `MISS` then `HIT`
2	Open dashboard (`http://127.0.0.1:9090/dashboard`)	Recent Traffic shows `miss` then `hit` on `/v1/chat/completions`
3	(Optional) Cursor Settings → Models → OpenAI Base URL = `http://localhost:8080/v1`	Chat traffic appears in dashboard

Common mistakes

Reading the chat reply — that is the model answer, not proxy logs.
Opening http://localhost:8080/v1/ in a browser — API only; shows BYPASS, not cache.
Using Kotro: Show Proxy Logs for HIT/MISS — that channel shows startup lines only; use Verify Cache or the dashboard.

Commands

Command	Description
Kotro: Verify Cache	Sends two identical test requests; confirms cache HIT
Kotro: Connect Cursor	Wizard for routing Cursor BYOK chat through the proxy
Kotro: Setup Continue.dev Config	Adds Kotro to `~/.continue/config.json`
Kotro: Open Dashboard	Opens the local operator UI
Kotro: Show Proxy Logs	Opens the extension output channel (startup / errors)

Architecture

IDE agent  →  localhost:8080/v1/*     (LLM proxy — may bind 0.0.0.0 in cluster mode)
Operator   →  127.0.0.1:9090/dashboard  (telemetry — loopback only by default)

Other install channels

npm: npm install -g @kotro-labs/proxy-engine
Homebrew: brew tap kotro-labs/tap && brew install kotro-proxy
GitHub Releases: kotro-labs/kotro-proxy-engine

Documentation

Full engine docs, threat model, and observability spec: github.com/kotro-labs/kotro-proxy-engine

License

MIT — Kotrosystems