Frona

Frona is a self hosted platform for running autonomous AI agents. You define agents, give them a task, and they work it out: browsing the web, running code, building and deploying applications, placing phone calls, delegating subtasks to other agents and carrying context between conversations. The engine is written in Rust and runs everything in a single process. It is aimed at people who want agent autonomy without handing an outside service their data or their credentials. Shell commands execute inside a sandbox that restricts filesystem access to the agent's own workspace, enforces execution timeouts and applies a network allowlist per agent, so an agent reaches only the domains you permit. Each agent gets its own tools, workspace and credentials, and browser sessions are isolated per user so cookies never leak between them. Credentials stay in your vault, with support for 1Password, Bitwarden, HashiCorp Vault and KeePass, and an agent must request each secret for you to approve or decline in real time; secrets are never written into agent memory or sent to a model provider. Hard CPU, memory and disk limits apply to every sandbox. Memory is ontology backed, preserving evidence and resolving identities so it stays searchable across conversations. You choose the model provider, and traffic goes from your own instance straight to it. Frona does not publish pricing. The documentation site carries no pricing page and no plan list, which is consistent with a product you deploy on your own infrastructure. The documentation marks the current release as a late 2026 version and includes a page comparing Frona against two other agent platforms on security, sandboxing and user experience.