OpenLegion

OpenLegion coordinates teams of AI agents that can use browsers and tools to carry out assigned work. Its templates include developer, sales and content teams, and users can define their own roles. The system is designed around separate containers, a shared coordination layer and budgets for individual agents. Those controls deserve evaluation alongside task quality, because an agent that can take actions needs clearer boundaries than a draft-only assistant.

Choose managed hosting or your own infrastructure

The product description offers a managed service with templates and an operator dashboard, plus a self-hosted path requiring Python 3.10 or later, Docker and model access. Its fleet model uses a blackboard, messaging and handoffs rather than one central decision-making agent. Each deployed container counts toward the agent allowance; a three-role developer template therefore consumes three agents.

The vendor describes browser automation, file handling and code execution. Start an evaluation with a bounded, reversible task and explicit permission limits. Claims that agents can perform any computer task without supervision are broader than the captured demonstrations establish. Customer savings and sales testimonials are vendor-published accounts, not a forecast for another business.

Examine the credential and budget boundaries

OpenLegion says agents call through a credential vault proxy rather than receiving API keys inside their containers. It describes non-root execution, resource limits and a per-agent permission matrix. Daily and monthly budgets are designed to stop model spending at a configured threshold. Those are architectural claims from the vendor, not an independent security assessment.

Credential isolation also does not eliminate the effects of an authorized action. A tool can send an email or change a file without exposing its key. Review the permissions, audit records and approval process needed for those actions before connecting business accounts. The page's references to model providers do not imply those companies endorse the service.

Separate hosting fees from model consumption

The captured pricing page displays Growth at $62 per month for five agents and five concurrent browsers, Pro at $152 for 15 agents and ten browsers, and Pro Max at $296 for 30 agents and 30 browsers. Promotional comparisons and a yearly selector are also present; confirm the checkout period and current price.

Paid plans include welcome model credits, which the vendor says do not expire. After using them, you can buy credits or supply your own provider keys and pay providers directly. The subscription is therefore not unlimited model usage. Cancellation retains access through the paid period, after which the VPS is deprovisioned and instance data removed. Export needed records before that deadline. The self-hosted engine is source-available under PolyForm Perimeter License 1.0.1; assess that license for your intended deployment rather than assuming unrestricted open-source use.