Agent Orchestration Studio

Coordinate AI agents from VS Code. Give the lead agent a task, let it involve specialists, and review their progress in the task and overview panels.

Agent Orchestration Studio connects to the model providers and agent runtimes you configure. It includes local task tracking and persistent memory, with optional Laya routing, embeddings, and external tools.

Features

• Team chat and task execution — use the @orchestrator chat participant, the Agentic chat view, or the Run Task command.
• Specialist agents — start with Lead, Coder, Researcher, Planner, Database, Reviewer, DevOps, and Documenter. Edit agents, assign different providers or runtimes, and control whether each can delegate.
• Delegation and plans — break larger requests into dependent or parallel tasks, follow hand-offs, and inspect task status and results.
• Provider choice — configure OpenAI-compatible or Anthropic model connections and choose a model globally or per agent.
• Tool controls — agents can use workspace tools and, when enabled, tools exposed by VS Code such as MCP tools. Tool access is filtered for each agent.
• Approval modes — choose whether workspace writes and persistent memory updates ask first, are approved automatically, or use Full access.
• Persistent memory — keep conversation summaries and episodic or semantic memories in a local SQLite database. JEV records entities and relations; optional vector search can improve memory matching when an embeddings model is configured.
• Optional routing with Laya — use a Laya SystemOne endpoint to help select agents, tools, and context. Disable it or use the rules fallback when Laya is unavailable.
• External agent runtimes — connect configured agents to Claude Code, Codex, or a custom CLI; an n8n webhook; or a persistent WebSocket agent service.
• Progress and history — inspect active work, tool activity, task results, related knowledge, and recent conversations in the sidebar and panels.

Screenshots

Example screens with demonstration workspace data.

Full screen with sidebar

Sidebar

Overview panel

Task workflow

Agent chat

Requirements

• Visual Studio Code 1.104.0 or later.
• A model provider endpoint and model to generate agent responses. Laya, embeddings, MCP tools, and external agent runtimes are optional.

Get started

1. Install Agent Orchestration Studio, then open Preferences: Open User Settings (JSON) from the Command Palette.
2. Paste these settings inside the existing outer { }. Replace the URL and model with values from your provider:

"agentOrchestrator.providers": [
  {
    "id": "main",
    "name": "My Provider",
    "protocol": "openai",
    "baseUrl": "https://YOUR_PROVIDER_URL/v1",
    "model": "YOUR_MODEL_ID",
    "models": ["YOUR_MODEL_ID"]
  }
],
"agentOrchestrator.activeProviderId": "main",
"agentOrchestrator.activeModel": "YOUR_MODEL_ID"

3. Run Agent Orchestration Studio: Configure Provider API Key, select My Provider, and enter your API key. It is stored in VS Code Secret Storage; you do not need to paste it into the settings file.
4. Run Agent Orchestration Studio: Open Agentic and send a prompt.

For an Anthropic-compatible provider, set protocol to anthropic. For a local OpenAI-compatible server, use its API URL for baseUrl. The extension does not supply a model or API service.

Optional Laya routing

Set agentOrchestrator.layaEndpoint to your Laya SystemOne endpoint, then run Agent Orchestration Studio: Configure Laya API Key and Agent Orchestration Studio: Test Laya Connection. Laya helps route requests and decide whether to retrieve memory, graph knowledge, or tools. Agent Orchestration Studio can use its configured fallback when Laya is disabled or unreachable.

Tools and approvals

The chat approval selector has three modes:

External VS Code tools can be enabled or disabled with agentOrchestrator.externalTools.enabled. Use agentOrchestrator.externalTools.toolPatterns, agentOrchestrator.externalTools.agentTools, and agentOrchestrator.externalTools.autoApprove to control which external tools agents can see and which may run without an approval prompt. Review tool permissions and approval mode before assigning work.

Memory and privacy

Conversation history, tasks, traces, memories, and knowledge graph data are stored in a SQLite database in VS Code's global storage for the extension. Embedding search is off by default. API keys and integration tokens configured through the extension's commands are kept in VS Code Secret Storage.

When an agent runs, the request and relevant context are sent to the model provider assigned to that agent. If enabled or configured, requests may also go to Laya, an embeddings endpoint, an external VS Code tool, an n8n workflow, a WebSocket agent, or a coding-agent CLI. Review those services' own data policies before use.

Configure external agents

Use Agent Orchestration Studio: Manage Agents to assign a runtime to an agent. Configure connections in agentOrchestrator.cliAgents, agentOrchestrator.n8nLinks, or agentOrchestrator.socketLinks. Store n8n and WebSocket tokens with Configure n8n Token and Configure Socket Token.

For WebSocket agents, the extension connects to wss:// endpoints (or ws://localhost) and sends a JSON message with type: "agent.execute", a requestId, task details, the agent, and the instruction. The service should answer with type: "agent.result" and the same requestId plus an output string, or type: "agent.error" and an error message. It may send agent.progress messages while working. A configured bearer token is included in the request's auth.bearerToken field.

Commands

Open the Command Palette and search for Agent Orchestration Studio.

Troubleshooting

• No model response: check the provider URL, protocol, model name, saved key, and network access.
• Laya is unconfigured: set agentOrchestrator.layaEndpoint and use Test Laya Connection, or turn Laya off to use the fallback behavior.
• An external tool is missing: check that the tool is registered in VS Code and matches the external tool settings and agent grants.
• A CLI agent is unavailable: install its CLI and ensure VS Code can find it on PATH.

Contributing

Contributions are welcome. See CONTRIBUTING.md for the development setup and pull request checklist. Please report bugs and request features through the repository's GitHub Issues page.

License

This project is licensed under the MIT License.