Skip to main content

Agent Runtime

An embeddable agent execution engine. Plug in any LLM provider, storage backend, and tool set to build production agents.

npm install @ainative/agent-runtime

Quick Start​

import { Agent, AINativeProvider, InMemoryStorage, webFetchTool } from '@ainative/agent-runtime';

const agent = new Agent({
  name: 'research-assistant',
  provider: new AINativeProvider({ apiKey: 'your-key' }),
  storage: new InMemoryStorage(),
  tools: [webFetchTool],
  systemPrompt: 'You are a research assistant. Use web_fetch to find information.',
});

const result = await agent.run('What are the latest trends in AI agents?');
console.log(result.output);
console.log(`Tools used: ${result.toolCalls.length}`);

LLM Providers​

Swap providers without changing agent logic.

import {
  AINativeProvider,
  OpenAIProvider,
  AnthropicProvider,
  OllamaProvider,
} from '@ainative/agent-runtime';

// AINative (default)
const ainative = new AINativeProvider({ apiKey: 'key' });

// OpenAI
const openai = new OpenAIProvider({ apiKey: 'sk-...' });

// Anthropic
const anthropic = new AnthropicProvider({ apiKey: 'sk-ant-...' });

// Ollama (local)
const ollama = new OllamaProvider({ baseUrl: 'http://localhost:11434', model: 'llama3' });

Storage Backends​

Agents persist memory and state across runs.

import { InMemoryStorage, LocalStorage, AINativeStorage } from '@ainative/agent-runtime';

// In-memory (development)
const mem = new InMemoryStorage();

// Local filesystem
const local = new LocalStorage({ path: './agent-data' });

// AINative cloud (production)
const cloud = new AINativeStorage({
  apiKey: 'key',
  projectId: 'my-project',
});

Built-in Tools​

ToolDescription
webFetchToolFetch and parse web pages
httpToolMake arbitrary HTTP requests
bashToolExecute shell commands
fileToolRead and write local files
gitToolGit operations (status, diff, commit)
mcpTools(config)Load tools from any MCP server

MCP Tool Adapter​

Connect any MCP server as agent tools:

import { Agent, AINativeProvider, mcpTools } from '@ainative/agent-runtime';

const agent = new Agent({
  provider: new AINativeProvider({ apiKey: 'key' }),
  tools: [
    ...await mcpTools({ command: 'npx', args: ['ainative-zerodb-memory-mcp'] }),
  ],
});

Dialog Manager​

For multi-turn conversations with context management:

import { Agent, DialogManager, AINativeProvider, AINativeStorage } from '@ainative/agent-runtime';

const agent = new Agent({
  provider: new AINativeProvider({ apiKey: 'key' }),
  storage: new AINativeStorage({ apiKey: 'key', projectId: 'proj' }),
});

const dialog = new DialogManager(agent);

// Multi-turn with automatic context
await dialog.say('My name is Alice');
const response = await dialog.say('What is my name?');
// "Your name is Alice"

Events​

Monitor agent execution in real-time:

import { Agent, AgentEventEmitter } from '@ainative/agent-runtime';

const agent = new Agent({ ... });

agent.on('tool:start', ({ tool, input }) => {
  console.log(`Calling ${tool}...`);
});

agent.on('tool:end', ({ tool, output, duration }) => {
  console.log(`${tool} completed in ${duration}ms`);
});

agent.on('llm:start', ({ messages }) => {
  console.log(`Sending ${messages.length} messages to LLM`);
});

Workspace Security​

The WorkspaceManager sandboxes file operations:

import { Agent, WorkspaceManager } from '@ainative/agent-runtime';

const workspace = new WorkspaceManager({
  rootDir: '/tmp/agent-workspace',
  allowedPaths: ['/tmp/agent-workspace/**'],
});

Next Steps​