Know the appliance is well.

The Dashboard is the operational starting point. It brings health, recent activity, and update state together so you can see the state of the appliance at a glance.

VERSION 0.1.20 · RUNTIME AND OBSERVABILITY

What the Dashboard shows

• System and appliance health — the current state of the services that make up Kautuka.
• Recent sessions — what has been running lately.
• Update state and history — which release track you are on and what has been applied.

A separate Usage view shows consumption per run and summarised over time, which is useful when a cloud model provider is metered.

Updates

Kautuka has two update tracks: stable, which is the default, and nightly/beta. Version 0.1.20 is the current stable release. Administrators choose the track. Changing track never downgrades you: switching to stable waits for a newer stable release rather than rolling back.

Applying or reverting an update involves a service restart, so expect a handoff and verify the reported version after the stack returns.

Advanced settings

Administrator-only settings under Settings → Advanced cover:

• Workspace — the effective workspace root, the default sandbox image, and CPU, memory, and timeout limits.
• Document ingestion — enablement, the parser connector or tool, timeout, extensions, PDF image fallback, and vision routing.
• Updates — the release track.
• Runtime — structured-output adaptation controls.
• Observability — system log level and optional raw-message capture.
• Persona, provider catalog, provider path, and memory defaults.

Raw-message capture records more of what passes through, which increases the exposure of sensitive data in logs and traces. Turn it on deliberately, for as long as you need it.

Files, sandboxes, and usage

Agent files are managed from the agent chat surface, and session assets belong to the session that produced them. Sandboxes are managed, read-only-root containers with configured resource limits. Whether an agent can use workspace tools at all is controlled by its sandbox setting.

When something looks wrong

Start with Dashboard health, and keep the exact error, status code, agent ID, and session ID. Then check the specific area:

• Ollama will not connect — use the provider’s Test, confirm Ollama is running and the model is in /api/tags, and confirm the address is reachable from Kautuka rather than only from your browser.
• An agent is missing or chat will not start — confirm a provider, a model, and an agent exist in that order, then select the agent. A saved agent that is missing from the workspace is a separate check.
• Connector tools are missing — test the connector, save its credentials, refresh discovery, enable it, then attach it to the agent and check that agent’s effective tool policy.
• A run waits after an approval — confirm the request moved from Pending to History, then reopen the related session. Do not submit a second decision.
• A channel does not receive messages — run the channel test, check ChannelHost health, the token or app permissions, the destination ID, the default target agent, and the sync mode.
• A scheduled run did not happen — see Schedules.
• Memory is absent — confirm fact extraction is on for that agent, complete a turn with an explicit lasting statement, and check the correct agent. Extraction is selective.
• 401 or 403 — a 401 means your session is missing or expired; sign in again. A 403 means you lack ownership or administrator access, so check your role instead of retrying.