HiveLang

HiveLang is Bothive's custom, AI-first domain-specific language (DSL) for defining capable agents. A HiveLang file describes an agent's instructions, allowed tools, and optional specialist sub-agents; your application supplies the AI provider and tool implementations.

Status

v5 is the version currently used by Bothive. It is an early, usable runtime; its API and language grammar may evolve before a stable 1.0 release. Do not expose untrusted HiveLang programs to high-privilege tools without a policy layer and explicit approval controls.

Relationship to Bothive

HiveLang originated as the custom DSL that powers agent definitions in Bothive. Bothive uses the same v5 language model—tokenizer, parser, runtime, tool-capability model, and hierarchical agent execution—that this repository publishes.

This repository is the open-source language runtime, not the Bothive hosted application. Bothive's deployment additionally supplies authenticated APIs, AI-provider adapters, integrations and OAuth credentials, data storage, authorization, billing, and production observability. Those application services are intentionally not included here. When Bothive adopts a new HiveLang version, the corresponding runtime should be released here as the next version of this package.

Install

npm install hivelang

Quick start

import { createRuntime } from 'hivelang';

const source = `
bot SupportAgent {
  instructions {
    You are a concise and helpful support agent.
  }
  capabilities { tickets.lookup }
}`;

const runtime = createRuntime(async (_systemPrompt, messages, tools) => {
  // Adapt this callback to your LLM provider's SDK.
  console.log(messages, tools);
  return { content: 'How can I help?' };
});

runtime.registerTool(
  'tickets.lookup',
  async ({ id }) => ({ output: `Ticket ${id}: open` }),
  {
    name: 'tickets.lookup',
    description: 'Look up a support ticket by ID.',
    parameters: {
      type: 'object',
      properties: { id: { type: 'string' } },
      required: ['id'],
    },
  },
);

runtime.loadCode(source);
const result = await runtime.execute('SupportAgent', 'Can you check ticket 42?');
console.log(result.output);

Core concepts

• instructions provides the system prompt for the agent.
• capabilities declares the tools an agent may call. Register each implementation in the host application.
• agent defines a specialist that can participate in hierarchical execution.
• createRuntime accepts an AI-provider adapter; HiveLang does not send requests to an AI provider itself.

See examples/devgenius.hive for a larger program.

Development

npm install
npm run check
npm test
npm run build

Security

Treat agent programs, tool descriptions, and tool results as potentially untrusted input. Keep tools least-privileged, validate tool arguments, and require application-level approval for consequential actions. See SECURITY.md.

Contributing

Issues and pull requests are welcome. Please read CONTRIBUTING.md and follow the Code of Conduct.

License

Apache-2.0. See LICENSE.