Skip to content

bmbouter/alcove

Repository files navigation

Alcove

Alcove

Sandboxed AI coding agents on OpenShift/Kubernetes — natively.

Alcove runs AI coding agents (Claude Code) in ephemeral, network-isolated containers with full auditability. Each task gets a fresh container, a scoped authorization proxy, and a complete session transcript. No persistent state crosses task boundaries.

Why Ephemeral?

Long-running AI agents accumulate context window contamination, credential exposure, and filesystem state that creates security and reliability problems. Alcove takes the opposite approach: one task, one container, then destroy it. See Problem Statement.

Security Principles

Alcove is built around five north-star security principles that ensure safe AI agent execution:

  • Safe Sandboxing — Every task runs in an ephemeral, network-isolated container that's never reused
  • Data Leakage Protection — Real credentials are never available to the LLM; Gate injects them at proxy level
  • Human-in-the-Loop Controls — Security profiles and approval gates ensure humans control what agents can access
  • Audit Records — Every network request and LLM interaction is logged with full session transcripts
  • Least Privilege — Gate's network proxy denies by default; only explicitly allowed operations are permitted

For implementation details, see Security Principles.

Components

Component Name Purpose
Controller Bridge Coordination, dashboard, REST API, scheduler
Worker Skiff Ephemeral Claude Code execution (k8s Job / podman run)
Auth Proxy Gate Network sandbox, token replacement, LLM API proxy
Message Bus Hail Task dispatch and status (NATS)
Session Store Ledger Audit trail, transcripts, proxy logs (PostgreSQL)

Status

Phase 1 complete with Phase 2 features underway. Working end-to-end pipeline: CLI, REST API, dashboard, task scheduler, ephemeral container execution (podman and Kubernetes), NATS messaging, PostgreSQL session storage, credential management, dual auth backends, skill/agent repos (Claude Code plugins loaded into workers), YAML task definitions (version-controlled reusable tasks from git repos), cross-repository workflow orchestration with per-repo credential validation, and per-task NetworkPolicy enforcement on Kubernetes. See Implementation Status for details.

Requirements

  • Go 1.25+
  • podman (rootless)
  • make
  • An LLM provider (Anthropic API key or Google Vertex AI credentials)

Quick Start

git clone https://github.com/bmbouter/alcove.git
cd alcove
make up
# make up auto-generates alcove.yaml with a random database encryption key
# Open http://localhost:8080 — log in with admin/admin
# Configure LLM credentials and providers in the dashboard

For the full setup walkthrough, see the Getting Started guide.

CLI Installation

One-Line Install (Linux/macOS)

curl -fsSL https://raw.githubusercontent.com/bmbouter/alcove/main/scripts/install.sh | bash

Windows (PowerShell)

iex (iwr -useb 'https://raw.githubusercontent.com/bmbouter/alcove/main/scripts/install.ps1').Content

Manual Download

Download platform-specific binaries from GitHub Releases:

  • Linux AMD64: alcove-linux-amd64
  • Linux ARM64: alcove-linux-arm64
  • macOS Intel: alcove-darwin-amd64
  • macOS Apple Silicon: alcove-darwin-arm64
  • Windows: alcove-windows-amd64.exe

Verify Installation

alcove version
alcove --help

Getting Started with CLI

# Connect to your Bridge instance
alcove login https://your-bridge-instance.com

# Submit a task
alcove run "Fix the bug in the login function"

# List recent sessions
alcove list --since 24h

# Stream logs in real-time
alcove logs <session-id> --follow

Documentation

Document Description
Getting Started Setup, configuration, and first task walkthrough
Configuration All environment variables, skill repos, task definitions
API Reference REST API endpoints, request/response formats
CLI Reference All commands, flags, and usage examples
Development Guide Building, testing, adding features
Architecture Component design, deployment diagrams, roadmap
Security Principles Five north-star security principles with implementation details
Architecture Decisions Resolved design choices (18 decisions)
Implementation Status Current state, what works, what is next
Problem Statement Why ephemeral agents
Credential Management How Bridge manages credentials and passes tokens to Gate
Auth Backends Authentication and authorization design
Changelog Release history and version notes

Contributing

See CONTRIBUTING.md for development setup, build instructions, and how to submit changes.

License

Licensed under the Apache License, Version 2.0.

About

Sandboxed AI coding agents on OpenShift/Kubernetes

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors