Alphora
A full-stack, async-first Python framework for building production AI agents, providing orchestration, tool execution, memory management, a secure code sandbox, a skills ecosystem and streaming in one OpenAI-compatible design.
Alphora provides ReAct and plan-execute reasoning loops with automatic tool orchestration, retry logic and iteration control; a @tool decorator that generates OpenAI function-calling schemas from type hints and docstrings with Pydantic V2 validation and parallel execution; and agent derivation, where child agents inherit an LLM, memory and configuration from a parent so a gateway can spawn orchestrators or specialists per request.
Memory supports multi-session isolation with a composable processor pipeline (keep_last, token_budget, summarize_tool_calls and similar), pinning and tagging, and undo/redo. Agent-generated code can run in Local, Docker or remote Docker sandboxes with file isolation and package management. Streaming uses typed async server-sent events, a Jinja2-based prompt engine supports parallel prompts and long-text continuation, and a unified, fail-open-by-default hook system covers tools, memory, the LLM, the sandbox and the agent lifecycle. Optional MCP integration and Agent Skills compatibility are installed as extras.