provision.ai

What It Is

Provision is an AI Agent cloud platform that hosts OpenClaw and positions itself around the idea of “hiring AI employees.” It gives each Agent a dedicated runtime, a sandboxed Chrome browser, a real @provisionagents.com email address, and one-click integrations with Slack, Telegram, Discord, and Web Chat. It is not a proprietary large language model product; instead, it packages Agent execution, browser access, email delivery, channel OAuth, and monitoring updates into a cloud service.

Core Capabilities and Integrations

Provision’s main strength is its engineering setup. Each team gets an isolated OpenClaw runtime, and each Agent has its own sandboxed browser for web navigation, form filling, screenshots, and similar tasks. Agents also get real email inboxes for sending, receiving, sorting, and following up on messages. Collaboration entry points cover Slack, Telegram, Discord, and web chat, making it suitable for bringing Agents into an existing team workflow. On the model side, it supports connecting existing ChatGPT Plus/Pro, Claude Pro/Max, OpenAI API, and Anthropic API accounts, or topping up Provision credits. The company says there is no markup when you bring your own tokens.

Pricing and Trial

The cloud version costs $299/agent/month, with a 48-hour free trial, no credit card required, and cancellation available at any time. The self-hosted version is MIT open source and free as software, but you need to cover your own server, LLM API, email, browser sandbox, and operations costs. For teams that want to avoid infrastructure maintenance, the cloud pricing is straightforward. For individuals or small teams, however, nearly $300 per Agent per month is a meaningful entry barrier, and a 48-hour trial may be a bit short for validating complex workflows.

Pros, Cons, and Limitations

Its advantages include a short deployment path, broad channel integrations, auditable open-source code, self-hosting support, and data export, all of which reduce vendor lock-in. It is especially suitable for multi-Agent scenarios such as marketing, SEO, SDR, research analysis, operations, and inbox handling. The limitations are that the available materials do not disclose details such as SLA, SOC2/ISO certifications, data residency, or enterprise permission systems. There are also no benchmark tests for output quality, so real-world performance will depend on the selected GPT/Claude/API model, prompts, and task design.

Who It’s For and Access from China

Provision is better suited to teams using an overseas collaboration stack, especially organizations already using Slack, Telegram, Discord, OpenAI, or Anthropic. Teams that do not want to maintain OpenClaw infrastructure but still want “AI coworkers with email and a browser” are likely to benefit. Chinese-language support is not clearly stated. Access from mainland China, payment availability, and model connectivity are not disclosed in the provided materials, so they should be considered unknown. If access or payment is restricted, alternatives include self-hosting OpenClaw, Dify, LangGraph, AutoGen, CrewAI, or n8n + LLM workflows.