Daemora launches as self-hosted AI agent with 14-layer security model

Daemora has launched as a self-hosted AI agent built around a security-first design, with its creator describing a 14-layer security model, AES-256-GCM encryption, subprocess isolation, and support for more than 25 model providers. The project is open source and aimed at users who want to run an agent on their own infrastructure rather than send their data to a hosted service.

The announcement comes from Dev Journal, where the creator describes Daemora as a local-first system that can act on behalf of a user while keeping tighter control over data flow and execution. That matters in AI agent software, where the agent is not just generating text, but can also call tools, access files, and trigger actions across a system.

According to the project description, Daemora ships with 52 built-in tools. Those tools give the agent access to common operations an automation system may need, such as interacting with external services, handling content, or moving between different steps in a workflow.

The security model is the main selling point. AES-256-GCM is a widely used encryption standard for protecting data at rest or in transit, while subprocess isolation means risky actions can be separated from the main process so a failure or compromise in one part does not automatically spread to everything else.

The creator also says Daemora uses a three-layer memory system. In AI agents, memory usually refers to how the system stores context about past interactions, user preferences, or task state so it can behave more consistently over time. A multi-layer approach typically separates short-term context from longer-term stored information, which can help keep the agent useful without forcing every detail into one bucket.

Daemora also supports more than 25 model providers, with automatic failover. In practice, that means the agent can switch between different AI models or vendors if one is unavailable, slow, or unsuitable for a given task. For teams building automation around external model APIs, that kind of fallback can reduce downtime and lower the risk of a single provider becoming a hard dependency.

Self-hosted agents have become more attractive as more organizations want to keep prompts, files, and workflow data under their own control. That is especially true for developers, security teams, and smaller IT shops that do not want every action routed through a third-party cloud service.

The security-first framing also reflects a wider problem in agent design. Once an AI system can read files, execute commands, or talk to services, the question is no longer just whether the model gives a correct answer. It is whether the surrounding architecture can contain mistakes, limit blast radius, and keep sensitive data from leaking across tasks.

Daemora’s launch positions it as an entry in that crowded middle ground between general-purpose chat tools and enterprise automation platforms. The project combines a large built-in toolset, model-provider flexibility, and a set of safeguards that are meant to make self-hosting more practical for people who need both automation and control.