Deploy and monitor autonomous AI agents with a few lines of code

Until now, building autonomous agents required engineering teams to construct complex infrastructure from scratch. This meant implementing and managing execution environments, agent loops, memory, tools integration, orchestration, and observability. This resulted in:

High costs requiring specialized AI engineering talent
Bespoke systems that are difficult to generalize or reuse
Limited visibility into agent behavior and failure modes
Rapid obsolescence as AI capabilities advance

SwarmKit SDK eliminates this complexity by delivering Autonomous Agent Infrastructure as a Service. With a few lines of code, deploy best-in-class AI agents that run in secure, isolated sandboxes with built-in observability and scalability.

import { SwarmKit } from "@swarmkit/sdk";
import { createE2BProvider } from "@swarmkit/e2b";

const sandbox = createE2BProvider({
  apiKey: process.env.E2B_API_KEY!,
});

const swarmkit = new SwarmKit()
  .withAgent({
    type: "claude",
    apiKey: process.env.SWARMKIT_API_KEY!,
    model: "claude-sonnet-4-5",
    reasoningEffort: "medium",
  })
  .withSandbox(sandbox)
  .withContextFiles([{ path: "rentroll.pdf", data: rentRollPDF }])
  .withSystemPrompt("You are an expert structured data extraction system and CRE financial analyst.");

const result = await swarmkit.run({
  prompt: "Infer the appropriate JSON schema for this rent roll, extract all tenant data, build a 10-year DCF model, calculate IRR assuming 6.5% cap rate exit, and save to Excel.",
});

const output_files = await swarmkit.getOutputFiles();

console.log(result.stdout);

Advanced Capabilities

Long-Running Agents with Memory

Persistent agents that can run 24/7 or on scheduled intervals, maintaining memory and context between executions. These agents continuously monitor data sources, detect patterns, and generate insights over time - ideal for applications like market trend analysis, thesis generation and live monitoring.

Wide Orchestration

Spawn thousands of agents in parallel for any task. This capability enables massive parallelization for applications like:

Large-scale research across hundreds of sources simultaneously
Processing thousands of documents in parallel for data extraction
Prototype or proposal sampling

Multi-Agent Systems

Leverage multi-agent patterns:

Executor-Reviewer: One agent executes tasks while another reviews, validates outputs and provides feedback
Coordinator-Subagents: A coordinator agent manages specialized subagents for complex workflows

Real-World Use Cases

VC THESIS GENERATION

Swarms of background agents that continuously scrape the web, analyze startup trends, and form early investment theses. These "slow hunch" engines gradually synthesize weak signals into actionable insights for venture capital partners.

STRUCTURED DATA EXTRACTION

Agents that automatically extract and structure data from complex documents (PDFs, spreadsheets, etc.), inferring appropriate schemas for CRE rent rolls, financial statements, and other domain-specific formats.