Companion is a key player in the environment and infrastructure layer of the AI agent stack. Rather than building models, they provide the necessary 'body'—a Linux sandbox with a file system and browser—that allows models to act as agents. Their focus on persistent memory and always-on execution addresses the primary limitations of current stateless LLM interfaces.
For the agent ecosystem, Companion acts as a production-ready host for the OpenClaw framework. They are championing the move toward asynchronous, multi-channel agents that can operate independently of a web tab. This makes them highly relevant to developers building autonomous researchers, personal assistants, or automated DevOps agents that need to survive session timeouts and work across different messaging platforms.
Companion is a cloud platform that provides AI agents with a persistent execution environment. Unlike standard large language model interfaces that reset after every session, Companion treats the agent as a tenant of a dedicated Linux sandbox. This allows the AI to maintain a file system, run a web browser, and execute code in the background without user supervision. The project is built on OpenClaw, an open-source framework developed by Peter Steinberger, designed to handle the orchestration and multi-channel routing required for autonomous agency.
The primary differentiator for Companion is its always-on nature. Most AI interactions are synchronous; the user sends a prompt and waits for a response. Companion agents are designed to work asynchronously. A user can assign a research task or a coding project on the web interface, close the browser, and receive a notification on Telegram or Discord once the work is complete. This persistence is backed by actual hardware allocations including CPU cores, RAM, and cloud storage. These resources give the agent a stable home where context and history are preserved across different platforms.
The platform is model-agnostic, supporting major providers like Anthropic, OpenAI, Google, and DeepSeek. This flexibility allows users to swap the brain of their agent while maintaining the same body—the Linux environment and its associated memory. For developers, this means they can focus on defining agent skills and custom instructions without managing the underlying infrastructure like networking, security, or compute scaling. By decoupling the model from the environment, Companion ensures that users are not locked into a single provider's ecosystem for their agent's operational capabilities.
Companion offers three tiers of service, starting at $40 per month. These plans are structured similarly to virtual private server hosting rather than traditional software-as-a-service seats. The Plus plan provides 2 CPU cores and 4 GB of RAM, while the Max plan scales up to 8 cores and 16 GB. Each tier includes monthly AI credits, though users also have the option to bring their own API keys. This pricing model suggests that Companion is targeting power users and developers who require dedicated resources for long-running autonomous tasks rather than casual chat users.
The company is led by Peter Steinberger, known for founding PSPDFKit and his contributions to the developer community. The startup, which recently raised $2.5 million in seed funding, is essentially a managed hosting provider for the OpenClaw framework. While the consumer AI market is crowded with companions designed for emotional support, Companion positions itself as a technical utility. It focuses on the execution of work that requires a computer's full suite of tools, from terminal access to file management. By hosting OpenClaw in the cloud, Companion removes the friction of self-hosting for users who want autonomous capabilities without the technical overhead of managing local sandboxes.
An always-on AI computer with persistent memory and a real Linux sandbox for autonomous tasks.
Companion is hiring.