Practical open-source tools for AI-native software and agent workflows.
HiAi builds focused tools for working with AI agents: shared knowledge, multi-agent development, portable plugins, observability, and reusable application infrastructure.
Most projects are designed to stay small, understandable, self-hostable, and easy to connect through APIs, MCP, and CLI tools.
Core workspaces, orchestration platforms, and observability stacks built specifically for agent-driven engineering.
Self-hosted AI-native knowledge workspace for people, applications, and agents. Rich editing, automatic embeddings, multilingual hybrid search, GraphRAG, REST API, TypeScript SDK, CLI, and MCP server.
Multi-agent engineering team for OpenCode. An orchestrator coordinates specialist agents for planning, implementation, review, research, design, and browser-based verification with enforced quality gates.
Lightweight all-in-one observability for applications and AI agents: errors, uptime, infrastructure, logs, OpenTelemetry, Sentry-compatible ingest, LLM tracing, MCP, CLI, and SDK.
Curated collection of portable Agent Plugins with trusted integrations, focused Agent Skills, safe defaults, and reproducible builds.
CLI for creating, migrating, and packaging portable Agent Plugins from declarative configs, existing agent setups, skills, and MCP servers.
Validation and diagnostics CLI for Agent Plugins. Checks compatibility and structure, explains problems, and safely fixes supported issues.
These repositories remain available as open source, but are not currently under active development.
Former standalone multi-agent coding environment. Development moved back to the plugin architecture in hiai-opencode.
HiAi projects are independent tools, not pieces of one mandatory framework. Where it makes sense, they share a few principles:
APIs, MCP, CLI, and agent-friendly workflows are first-class features.
Avoid unnecessary infrastructure and abstraction.
Important data and services should be able to run on infrastructure you control.
Avoid locking workflows to one model, agent, or vendor.
Projects should remain useful independently and integrate cleanly when used together.
Explicit configuration, automated validation, and predictable builds.
Public projects use MIT or Apache-2.0 licensing.