Skip to main content

Golem Cloud

Golem Cloud provides a durable runtime for AI agents, ensuring they survive crashes, redeploys, and long idle periods without losing state or re-executing tool calls. Built on WebAssembly, it runs each agent in an isolated VM with permission checks that AI-generated code cannot bypass, and supports six SDKs across five languages. The platform is designed for production-grade agents in coding, research, support, and business automation.

Traction 7/10
New York, United States · HQ
Founded 202381K+ followers
  • AI Agents
  • Artificial Intelligence
  • Developer Tools
  • Software Only
Updated 15 days ago

Funding

Funding rounds are not available yet.

Founders

Founder details are not available yet.

Product

Problem

AI agents in production face reliability challenges including state loss during interruptions, duplicate tool executions, and security vulnerabilities from prompt injection. Developers must build complex infrastructure for persistence, exactly-once execution, and policy enforcement, which is error-prone and time-consuming.

Solution

Golem Cloud provides a durable agent runtime that automatically persists state and effects without serialization or annotations, allowing agents to suspend for days or weeks at zero compute and memory cost. The platform executes code transactionally, ensuring tool calls complete exactly once even through crashes or restarts, while each agent runs in its own WebAssembly VM with isolated filesystem, database, and environment variables. Policy enforcement is built-in through explicit grants for tools, network, and file access, with runtime-enforced quotas and journaling of all grants. Developers can write agents that pause for human-in-the-loop approvals via webhooks, resuming seamlessly when approval is granted. The runtime also supports deploying agents, tools, and interactive artifacts to a single durable environment, with streaming methods and resumable token streams for browser-based output.

Target Audience

Primary customers are software developers and engineering teams building production-grade AI agents and autonomous systems that require reliable state management, exactly-once execution, and strict policy enforcement, particularly those creating coding agents and business automation workflows.

Features

  • Automatic state persistence with no serialization, state machines, or annotations required
  • Transactional execution ensuring exactly-once tool calls through any interruption, including crash mid-execution
  • WebAssembly VM per agent with isolated filesystem, SQLite database, and environment variables, with millisecond startup
  • Explicit grant system for tools, network, and file access with runtime-enforced quotas and journaled grants
  • Zero-cost suspension for long-running workflows, with agents resuming with the same memory, locals, and call stack
  • Built-in human-in-the-loop approval via webhooks that can wait for days at no cost
  • Six SDKs across five languages (TypeScript, Effect, Rust, Go, Scala, MoonBit) sharing one agent model
  • Durable Streams protocol for resumable token streaming directly to browsers
  • Read-only methods with cache policies (no caching, cache-until-write, TTL) enforced at runtime
  • Reflection capability allowing agents to discover and invoke other agents with typed clients
  • HTTP routers, static and live file serving, and front-end authentication for deploying interactive agent artifacts
  • Bidirectional MCP (Model Context Protocol) support for tool interoperability
  • golem ssh command for secure agent access
This profile is AI-generated and may contain inaccuracies.