Ask TIDE

The console’s question service, and its provisioning chat. See also Ask.

What it does

You ask a question in plain words. Runink TIDE looks it up with its own tools and answers only with figures it can point to.

• Intelligence questions: CapEx findings and their traces, the rule book, remediations, playbook runs, lineage, the systems inventory and data-governance findings.
• DataEx questions: model cards, the agent roster, autonomy settings, the harness findings queue, judgements and inference status.
• Provisioning chat: from a sentence, it proposes one action, such as listing instances or creating an instance for an app and tenant.

What it reads

Only what TIDE already holds, through read tools, one per step. Each question is scoped to Intelligence or DataEx, and the tools available depend on that scope.

What it produces

• A streamed answer in which every figure appears verbatim in a tool result, cited to a record that holds that value.
• A clear “could not look” with the reason, when a source cannot be read, rather than “nothing found”.
• For the chat: one proposed action, waiting for confirmation.

Human oversight

A person decides what to do with an answer. Every state-changing chat action needs a person’s confirmation before it runs. Ask acts for the signed-in person, and refuses on a console with no sign-in.

Model

Qwen3.6-35B-A3B, by Qwen, licensed Apache-2.0 (the general tier). Runink domain adaptation for this agent is planned; this release uses the base model.

Where it runs and data handling

On your TIDE deployment’s own inference, on your Server or in your cloud. Questions and tool results go to that model plane and to no third-party AI service. When the deployment’s console is set up to sign its model requests, a person’s question is queued ahead of background agent work.

Guardrails

• Figures are checked by the server: a number that is not verbatim in a tool result is removed from the answer.
• It is instructed never to produce confidence scores, quality ratings or personal-data judgements of its own.
• Tool results are untrusted data, and the model is told to ignore instructions inside them. Questions and answers are screened by the guardrails. Its tools only read; a chat action changes nothing until a person confirms it.
• An Ask question has a length limit, and each run a step and time budget.

Limitations

• It does not compute new figures, sum or convert. It reports what the systems hold.
• It does not answer delivery (DevEx) questions.
• When inference is not configured, it says so and calls no model.

Evaluation

No published evaluation scores yet.

Illustrative example

Invented records. “Why is project P-114 flagged?” Ask TIDE looks up the project’s findings, then traces one. Answer: “P-114 has 3 open findings. F-2231 (unit price above threshold) traces to purchase order PO-77812, line 4, unit price 812.00.” Every figure in the answer came from a tool result.