v0.1.0

Atelier v0.1.0 — Release Notes

The first public release of Atelier — an open-source platform for building, deploying, and orchestrating AI agents with tools, memory, and event-driven workflows.

Highlights

• Agent Builder — create and configure AI agents with a visual editor, attach tools, skills, prompts, and sub-agents
• Event-Driven Workflows — external apps emit events (e.g. todo.completed); Atelier routes them to matching agent workflows automatically
• MCP Support — connect to any Model Context Protocol server (HTTP, SSE, stdio) and give agents access to external tools
• Multi-Provider LLM — supports Google Gemini, OpenAI, and Anthropic models out of the box
• Workspaces & Teams — multi-tenant with RBAC (owner/admin/member), workspace-scoped resources, and invite links

Core Features

Agents

• Agent CRUD with streaming chat, session history, and memory
• Multi-agent orchestration with sub-agent composition
• Agent-to-Agent (A2A) protocol support for remote agent communication
• Configurable system prompts, instructions, and model selection

Built-in Tools

• call_api — make HTTP requests to external APIs
• web_search — search the web via Google (ADK built-in)
• send_email — send email via Gmail SMTP with encrypted App Passwords
• notify — push in-app notifications to the user
• schedule — create cron-scheduled or one-time agent runs
• workflow — create/manage event-driven workflows from within an agent
• events — register event types and query subscriptions
• self_learning — agent saves conversation patterns as reusable skills
• ui — render rich cards (plans, checklists, info boxes) in chat
• run_python — execute Python code with auto-dependency installation
• bash — run shell commands in a sandboxed workspace
• File tools: read_file, write_file, edit_file, glob_files, grep_files
• Math tools: add, subtract, multiply, divide, power, sqrt, percentage
• Time tools: get_current_time, get_current_date
• Database query tool with schema introspection (read-only option)

Skills

• Custom skill system with file tree editor (Python, shell scripts)
• Isolated execution in persistent virtual environments
• Auto-dependency installation on ModuleNotFoundError
• Skills attach to agents and provide domain-specific tools and instructions

Event-Driven Workflows

• Apps emit events via POST /api/v1/events with type + payload
• Workflows subscribe using glob patterns (todo.*, *.created, *)
• Per-workflow execution mode: serial (one at a time) or parallel (configurable concurrency)
• Priority-based queue with exponential backoff retry (up to 3 attempts)
• Event type registry with optional JSON Schema payload validation
• Global queue view with status filters (pending/running/done/error)

Schedules

• Cron-based recurring agent runs
• One-time scheduled executions
• Background scheduler daemon with automatic execution

Memory & Knowledge

• LLM-powered fact extraction from conversations
• Semantic memory storage in PostgreSQL
• Per-agent memory with search an...