FAQ
Questions that cut across the sections. Each answer links to the page with the detail.
General
Is Runink TIDE the same as Kubernetes?
No. TIDE runs on k0s and deploys with pure kustomize, without Helm. It adds its own operators, the
ClientInstance tenant API, the security spine, the agent fleet and the console. See
Architecture.
Do the apps depend on TIDE?
No. FACE, PULSE and LUNA do not import TIDE, and neither does the service behind TIDE’s FORGE
category. Each carries a test that fails if it does.
Code more than one of them runs lives in shared libraries (store, ui, inference, mesh,
security, ml).
How do I sign in?
With Google (when GOOGLE_CLIENT_ID is set) or with a username and password (when
CONSOLE_ADMIN_PASSWORD is set), at your installation’s own console URL. Your address must be on
CONSOLE_ALLOWED_EMAILS or in CONSOLE_ALLOWED_HD: with neither set, every Google sign-in is
refused. There is no Microsoft or other identity provider in the console code. See
Sign in.
Which models does TIDE use? Does it call a hosted AI service?
It calls no hosted model. The inference plane is sovereign, and mistral.rs serves every language tier: general and coder (Qwen3.8-27B) from an inference server outside the cluster, the smaller tiers from that server or the cluster.
Every model call goes through cmd/modelrouter, the one admission queue in front of the engines. See
Models and inference.
Does TIDE send my data anywhere? TIDE dials a tenant source in one place only: DataEx › Estate, and only when an admin runs one of its four actions (Test, Explore, Read access patterns, Map estate). Reads never dial. Agents registered in the Atlas workspace are metadata: TIDE never calls, polls or sends data to one. See Estate.
How is TIDE priced? TIDE is sold on its own, in three categories: Lite on Runink’s shared hardware ($85 a seat a month yearly, or $99 month to month, with 18,000 CU a seat), Dedicated on runners in your own cloud account ($279 a month for the first node, then $169 for each extra node, unlimited CU, plus 1.5% of the runners’ cloud cost), and Enterprise (custom, and the only plan with on-prem runners). FORGE is part of TIDE. One CU is one second of a model slot, measured by the model router. See Plans and billing.
Where did Argo CD go?
It was removed. Delivery is the tide-cd.yml GitHub Actions workflow, and DevEx › Clusters › GitOps compares
what it delivered with what is running. See Clusters.
DevEx
How long does a coding session take?
Budget in decode time, not wall-clock intuition. On CPU-only hardware the coder tier decodes a few
tokens per second, so a 40-turn session is roughly an hour. tide session start defaults to
--max-turns 40, --timeout 2h and --turn-timeout 15m. See Coding sessions.
Can I stop a running session?
There is no tide session stop. A run ends when it finishes, or at --max-turns, --timeout or
--turn-timeout. Sessions persist under ~/.core/sessions/<id>, and tide session resume <id>
continues one with a new instruction.
Why does the curator’s daily run succeed but publish nothing?
The curator is dry unless the Actions variable CURATOR_CRON_DRY_RUN is the literal string false.
Unset means dry, and a dry run finishes green. See Agents.
Why are reviews slow and one at a time? The engines decode a bounded number of requests at a time and the model router queues the rest, so reviews wait their turn. See Reviews.
DataEx
Does TIDE scan my data every day? The data governance agent runs daily, but it reads no rows and no values. It assesses data quality and PII exposure only from the structure and counts that Estate’s Explore recorded. A source nobody explored is reported unassessable, with the reason. See Data governance.
Why does my runner show offline when it is connected?
Estate actions and CapEx pulls run on the Runink managed runner (console). An action routed to a
connected self-hosted runner is answered runner_offline, with the detail “connected; …”, and
nothing runs. Choose console for the action. See Runners.
What is the difference between findings and judgements? A finding is a claim made by an agent or an assessment platform. A judgement is TIDE’s own verdict on a finding that an external platform submitted: concur, dissent, or unable to judge. Unable-to-judge is never shown as agreement. See Agents.
Why does the judge always report unable to judge?
Its docket can only be filled through POST /api/judgement/submissions, which needs
TIDE_JUDGEMENT_INGEST_TOKEN. Until an admin issues that token to a submitting platform, the docket
stays empty and the judge says so rather than guessing. That is the design working, not a fault.
Intelligence
Where do the Intelligence pages come from? They are a page-for-page Flutter port of the partner product Atlas, over TIDE’s own data and TIDE’s own in-process gRPC-web services. See Intelligence.
Why are most rules “Shadow” in a new workspace? The rules reconciliation calls a rule Shadow when the engine runs it but no governance decision declares it. In a fresh workspace nobody has decided anything yet, so most rules are Shadow. See Business rules.
FORGE
Is FORGE a separate site?
No. FORGE is a category of the TIDE console: the Studio is a console route (?tab=forgeStudio), under TIDE’s sign-in. There is no FORGE domain and no
FORGE login. See FORGE.
Who runs CI for a forged repository?
TIDE does, centrally. Forged repositories carry no CI of their own and no App credentials. TIDE
posts the result as the core/ci commit status. See Central verification.