ainclave runs autonomous coding agents in isolated microVMs where network egress is closed by default. Each agent session gets its own Firecracker microVM with its own guest kernel, so there's no shared kernel namespace between sessions. You point an agent at your git repo, bring your own model token from any provider, set optional runtime caps, and let it work without babysitting.

It's built for developers and teams who want to automate code tasks (writing, testing, refactoring) without the operational overhead of self-hosting or the security blind spots of shared cloud runners. You keep full control: your token stays encrypted, your repo access is scoped, your agent's network access is explicit and auditable.

Key capabilities:

• Isolation: Every session runs in its own microVM on bare metal at an EU provider, not in a shared container or tmux session.
• Closed egress: Network access is denied by default. An agent only reaches destinations you configure in its session setup.
• Bring your own token: Use Claude, GPT, or any model you choose. Your API keys are encrypted and never displayed after saving.
• No runtime ceiling: A single run can take hours or days if needed. You set optional caps per account and per session, both with auto-stop.
• Live audit feed: Watch what your agent does in real time. Boundary decisions (network access attempts) are logged as OCSF records.
• No surveillance: The platform uses guardrails, not content monitoring. You control what the agent sees and reaches.

The trade-off between control and convenience is real in most platforms. ainclave tries not to accept it: you get full control over agent permissions and visibility without managing infrastructure yourself. Standard git flow, no lock-in, no runners to keep alive.