Alpha 5: Agent-Centric Platform, Sandboxed Execution, Multi-Provider

Alpha 5 is the biggest release since I started Apprentice. It's now a true agent runtime, not a workspace tool. Agents are the center of everything: they own their runs, memory, budgets, and tools.

What's new

🧠 Agents are first-class
Create, enable, disable, and delete agents with a guided template gallery. Each agent has its own identity, model, prompts, budget, and permissions, and runs independently.

📦 Every agent runs in its own Docker sandbox
Agents can't touch anything you haven't explicitly allowed. Each one gets an isolated Linux user, its own folders, and a clean network boundary. You can safely let them execute code, read files, and use the web.

🤖 Multi-provider support
Bring your own Claude subscription, or run fully local with LM Studio. Switch providers per agent. No vendor lock-in.

💰 Budgets and execution limits
Set daily, weekly, or monthly caps per agent, plus global execution limits across your whole setup. Agents automatically pause, notify, or hard-stop when they hit a limit.

🛡️ Guardrails and permissions
AI-moderated safety rules, allowed-directory lists, website filters, and a permission broker. Credentials are stored in your OS keychain.

🗓️ Scheduling and triggers
Run agents on cron schedules, from webhooks, or from incoming messages on Telegram, Discord, Slack, Signal, WhatsApp, and Email.

💬 Group chat and inter-agent collaboration
Agents can talk to each other, mention each other, and share topics, like a Slack workspace for your AI team.

📚 Knowledge base and memory
Give agents long-term memory with local semantic search. No data leaves your machine.

🖥️ Isolated browser per agent
Agents browse the web in their own scoped sessions. Credentials and cookies stay isolated per agent, and you choose which ones to share.

📊 New Dashboard
A single overview of all your agents, their health, alerts, costs, and activity.

🔌 Richer tool ecosystem
Built-in tools for knowledge, memory, messaging, notifications, scheduling, tasks, activity, collaboration, and more, all via MCP.

🔔 Desktop notifications
Get native OS notifications when your agents need attention, even when Apprentice is in the background.

💳 Pro license
Pro license support is live with Community and Pro tier agent limits and capability toggles.

Under the hood

• Unified agent execution queue with concurrency control and stale-run recovery
• Graceful shutdown that waits for in-flight work
• Consolidated database bootstrap for clean installs
• Audit logging for agent runs and integration polling
• Cross-platform builds for Windows, macOS, and Linux

Known limitations

Alpha software. Expect rough edges, breaking changes before v1.0, and the occasional bug. Please report anything broken in Issues.

Get it

Download for your platform below. You'll need Docker Desktop (Windows and macOS) or Docker Engine (Linux) installed.