Claude Code Integration

Overview

Aileron currently provides its most complete agent experience through Claude Code. This gives teams a ready-to-use way to run Claude Code agents directly from the browser UI for code generation, analysis, refactoring, and day-to-day development work.

Aileron is not limited to Claude Code, though. The platform is evolving toward a broader multi-agent architecture, with OpenCode, Gemini, and Codex support being expanded over time. For now, Claude Code remains the most complete and production-ready integration in the product.

The goal is not only to integrate a powerful agent, but to make that agent usable inside governed, repeatable, enterprise-ready workspaces.

Agent Support Status

Agent	Status	Notes
Claude Code	Fully supported	Most complete today: chat, settings, automation, and OpenSpec workflow integration
Gemini	Partial support	Settings and foundational integration exist; end-to-end support is still expanding
OpenCode	Partial support	Settings and foundational integration exist; more workflow and execution support is planned
Codex	Partial support	Settings and foundational integration exist; more execution and governance support is planned

Features

Agent Session

Each Claude Code invocation creates an Agent Session:

Frontend Chat Panel
  │  Create session
  ▼
Workspace Runtime
  │  Run claude CLI
  │  Stream output (stdout/stderr)
  ▼
WebSocket → Frontend
  │  Real-time display of thinking / tool use / output

Supported Permission Modes

Mode	Description
default	Standard mode; tool calls require confirmation
acceptEdits	Auto-accept file edits
dontAsk	Never prompt; auto-execute all operations
auto	Fully automated mode

Claude Code Settings Management

The workspace settings page lets you manage:

• Model selection: choose the Claude model
• Hooks: configure custom pre/post hook commands
• MCP servers: configure MCP server connections
• Sub-agents: Claude Code sub-agent settings
• Slash commands: custom / commands
• Environment variables: inject ANTHROPIC_API_KEY and other runtime env vars

Environment Variable Injection

The Claude API key and other sensitive settings are configured dynamically via the frontend "Environment Variables" page and injected into the workspace runtime container. This helps platform teams avoid baking environment-specific secrets into images or local setup steps.

Chat Panel

The frontend Chat Panel provides:

• Streaming output: real-time display of Claude's thinking process and tool calls
• Tool-call visualization: shows file read/write, bash execution, and other operations
• Permission mode toggle: switch authorization modes mid-conversation
• Session management: view session history, resume, or start new sessions

WebSocket Agent Session

// Connect to Agent Session WebSocket
const ws = new WebSocket(
  `ws://localhost:3002/api/v1/ws/agent-sessions/${sessionId}`
);

ws.onmessage = (event) => {
  const msg = JSON.parse(event.data);
  // msg.type: 'thinking' | 'tool_use' | 'tool_result' | 'text' | 'done' | 'error'
};

Automation Integration

Claude Code can be scheduled to run via the Automation feature — see Automation Tasks. Automation is also most complete with Claude Code today, while broader multi-agent scheduling support continues to evolve.