Agents

Agents with an identity, rules and an audit trail.

Every agent is a client of the same API as the dashboard and the CLI. Writes, deploys, shell and secrets pass an approval gate; every turn is an event you can read back.

Docs
xhigh · /effort
auto mode on (shift+tab to cycle) · ← for agents

The question

Can an agent take production down?

Not without a person. Sensitive actions wait at an approval gate, changes to installs go through reviewable requests, and the agent’s identity, skills and grants decide what it may touch at all.

What is in it

Agents, piece by piece.

What each part does, and whether you can use it today.

Code Mode

Beta

akua.request() typed bridge with server-injected auth.

Agent identities

Soon

Durable configured persona + capability bundle.

Agent sessions

Soon

Workspace-scoped conversation / task instances.

Agent turns & events

Soon

One submitted instruction; append-only normalized stream.

Approval gates

Soon

Human approval for writes / deploys / shell / network / secret access.

Agent skills

Soon

Versioned reusable behavior modules; required-grant catalog.

Dashboard agent

Soon

Embedded chat with widget-rendering events.

Ambient agents

Soon

Continuous watchers that proactively open investigations.

Hosted coding sandboxes

Soon

Codex CLI / Claude Code / Akua CLI inside isolated runtimes.

What it costs

The platform is a flat fee. Compute is at cost.

Free to start with one managed cluster and one machine. Pro is €20 per workspace and month. Your own cloud key removes every limit, and Akua never marks up compute.

See the plans

Start with agents
on your own terms.

Free to begin, on Akua-managed machines or your own cloud key.

Docs