vLLM Studio

Unified local AI workstation for model lifecycle, chat/agent workflows, orchestration, observability, and remote deployment.

Lite Mode (Agent Desktop)

For users who only need the coding agent without managing vLLM infrastructure:

• Default ON - shows only Agent + Settings tabs
• Toggle in Settings → Appearance → Interface Mode
• Switch to "Full (vLLM Infra)" to access Status, Usage, Models, Server tabs

Use Lite Mode when your org manages the vLLM deployment and you just need the agent desktop.

OpenRouter Support

Connect to OpenRouter or any OpenAI-compatible API:

1. Go to Settings → Connection
2. Set API URL: https://openrouter.ai/api
3. Set API Key: your OpenRouter key
4. Save and select model in Agent dropdown

Works with any provider exposing /v1/models and /v1/chat/completions.

Mobile Panel

Test mobile apps directly from the agent workspace:

• Device control: list, select, boot emulators/simulators
• Live view: iOS streaming via serve-sim, Android via screenshot polling
• Touch input: tap-to-interact on device screen
• Hardware buttons: HOME, BACK, etc.
• Logs: device log viewer

Requires mobilecli (npm) for device control. Optional serve-sim for iOS 60fps streaming.

npm install -g mobilecli
npm install -g serve-sim  # optional, iOS only

Agent Verification (built-in)

The internal coding agent ships with a verification loop baked into the runtime — no .cursor/skills/ setup, no opt-in flag, no per-project rules. Every fresh install gets it.

After each file edit the agent is instructed (and, if needed, forced) to verify the change against the running app:

• verify_web — navigate the embedded browser to a URL and capture screenshot / DOM / a11y tree.
• verify_mobile — screenshot a connected device (iOS sim or Android emulator) and optionally tail logs.
• verify_responsive — capture the same URL across mobile / tablet / desktop viewports.
• verify_until_pass — iterative grind loop (up to 3 iterations) whose transcript the agent self-evaluates against free-form success_criteria.

Two enforcement layers ship together:

1. A built-in system-prompt addendum is injected on every agent spawn, telling the model when to call each tool.
2. A server-side safety net in /api/agent/turn watches the SSE stream — if the agent edited frontend or mobile files but never called a verify_* tool, the server automatically issues a follow-up prompt that forces verification before closing the turn. Look for auto_verify SSE events in the chat stream.

A user-facing toggle to disable auto-verify is on the roadmap; today it is always on.

Release: v1.13.0

This release consolidates major repo changes currently in the tree, including:

• OpenAI proxy activation policy controls for load_if_idle and switch_on_request
• lifecycle-aware run aborts when model eviction happens
• SSE run stream termination fixes across backend and frontend
• local-only chat/runtime cleanup and controller simplification
• dashboard launch-state cleanup improvements
• reduced chat/controller indirection and removed dead remote-runtime branches

Docs

• Overview: docs/README.md
• Setup and deployment: docs/operations.md
• Environment variables: docs/environment.md

Repository layout

• controller/: Bun/Hono backend, orchestration, chat runtime, lifecycle, metrics
• frontend/: Next.js app, chat UI, proxy endpoints, client state
• cli/: Bun CLI for controller access
• shared/: shared types/contracts
• config/: runtime and integration configs
• docs/: documentation index and environment notes
• scripts/: operational scripts (deployment + controller daemon helpers)
• docker-compose.yml: full stack service definitions
• scripts/daemon-*.sh: start/status/stop helpers for background controller runs

Quick start

1. Controller (local):

cd controller
npx tsc --noEmit
bun test
bun src/main.ts

2. Frontend:

cd frontend
npm run test
npm run lint
npm run build
npm run dev

3. Full stack with Docker (controller + frontend + infra):

docker compose up -d --build controller frontend

4. Run controller as a background daemon:

./scripts/daemon-start.sh
./scripts/daemon-status.sh
./scripts/daemon-stop.sh

Health checks

curl -sS http://localhost:8080/health
curl -I http://localhost:3000

API docs

• http://localhost:8080/api/docs
• http://localhost:8080/api/spec

Setup guide

See docs/operations.md for setup, deployment, and verification instructions.

Branching and release workflow

• Development branch: dev
• Production integration branch: main
• Release tags: vX.Y.Z

For this release:

• merge release work into main and dev
• tag v1.13.0
• create a new post-release working branch