HiveLang
A declarative language and execution runtime for tool-using agents, specialist sub-agents, policies, hooks, and workflows.
The Story
HiveLang is an AI-first DSL for describing agent behavior. A bot definition can declare instructions, capability names, grounding and memory policies, hooks, specialist agents, schedules, webhooks, and reasoning handlers.
What I Built
Source is tokenized and parsed into an AST, validated for configuration errors, then loaded into a runtime that builds bot configuration and maps declared capabilities to registered tools. The runtime accepts an AI-provider adapter rather than hard-wiring one model vendor. The v5 runtime keeps its own execution context for variables, outputs, errors, tool calls, memory, and checkpoints. Specialist agents retain distinct system prompts and tool subsets rather than being flattened into the parent agent. The runtime guards reasoning iterations and delegation depth, and its optional collaboration hooks are best-effort so persistence failures do not terminate an agent run. Parser diagnostics also catch invalid configurations before execution.