clio-agent

CLIO Agent is an autonomous AI system designed for managing large-scale scientific datasets. It operates as the Intelligence Layer (CEI) of the IOWarp platform, providing a three-tier architecture consisting of a main orchestrator (planner loop over registered tools and experts), domain-specific experts (data, analysis, visualization), and ephemeral nanoagents for parallel sub-tasks. All tiers share state in a local-first ARC memory layer enabling fast context retrieval with O(log N) complexity. Key capabilities include multi-expert capability-based routing, a FastMCP tool gateway that natively handles HDF5, Parquet, and CSV formats, and an optimizer layer for both offline tuning and online learning. A runtime doctor monitors LM, gateway, and file-policy health, while the system works with diverse LLM backends including Ollama, OpenAI, Anthropic, and OpenRouter. Users interact via a CLI (`clio`) and a Bubbletea terminal UI (`clio --web`), with deployment options ranging from source builds and Docker containers to pre-built desktop applications and web bundles. The agent follows a modular design where experts are registered blueprints that can be composed into plans. The orchestrator dynamically routes tasks among experts based on their capabilities, and the optimizer layer enables continuous improvement of workflows. Configuration is flexible through environment variables such as CLIO_PREFIX, CLIO_BIN_DIR, and CLIO_PORT, allowing customization of paths and server ports. The system is BSD-3-Clause licensed and supported by the Gnosis Research Center at Illinois Tech.