Know the system.
Shape what comes next.

DecaCap is a configurable multi-model Workbench. Each entitled box can use its own primary role, support role, provider, model identifier, task, and route. Solo Board Chat can target selected boxes, while Solo Begin launches all enabled entitled boxes in parallel. Collaboration runs sequential worker handoffs with a supervisor review between them; Dual divides the workers between two reviewing supervisors.

No. Obmil identities stay attached to their boxes, but functional roles do not. You can mix and match primary roles, support roles, specializations, providers, and models for every project.

Spores provides quality guidance and suggested role combinations. Magic contains Solo, Collaboration, and Dual project prompts that can inject suitable roles, support roles, rules, tasks, and outlines into the Workbench for you to review before running.

No. DecaCap uses BYOK: each user chooses a supported provider, supplies that provider’s key, and pays any provider usage charges directly. A local-only workflow does not require a cloud-provider key.

No. A cloud-provider connection is entered once. Every higher-numbered box assigned to that same provider reuses the encrypted connection while keeping its own model identifier, role, and task configuration. Local Model uses the device companion instead of a cloud API key.

Some providers do not expose a compatible general model endpoint, or require a specialized media adapter. DecaCap keeps those paths visible for workflow planning but labels their execution boundary instead of pretending they work.

Not because a fixed amount of time elapsed. Worker requests have no DecaCap execution cutoff. In Solo mode, only the user can stop a box. In Collaboration and Dual modes, only the assigned Majesty Obmil Queen supervisor can stop or take over a worker—either through its visible Stop active control or through an enabled timed check that finds explicit evidence in streamed output matching the user’s stop or takeover rules. Silence, slow output, and non-streaming Kimi wait time are never stop conditions.

The user does. DecaCap does not own a central GPU or model. The free DecaCap Local Companion connects this browser to a model runtime on the user’s own machine, and the model files, prepared data, adapters, and training output stay there.

Yes. Any entitled box can choose Local Model as its provider and use the same paired device companion and runtime-model name. The request goes from that normal Workbench box to the user’s own companion instead of through DecaCap’s server. GoldCap is not the only local-capable box; it is the dedicated training hub.

GoldCap is the only destination for training documents and local training controls. It separates data preparation from training, can capture user-approved prompt/output pairs, accepts supported structured or readable documents, prepares local training files, launches a compatible local fine-tuning command, reports the process state and any progress visible in its logs, pauses or resumes where the operating system supports it, and stops the process. Its isolated test targets the runtime model selected by the user; DecaCap does not automatically install or activate a newly created adapter.

Yes, while the Workbench page and local companion are both running, after you explicitly enable automatic pair preparation in GoldCap. Completed pairs are delivered to the local companion. If automatic training is also enabled, a compatible configured profile can start at the threshold you choose. Both options are off until you approve them.

Cloud-provider keys are encrypted in the user’s private account vault and are never returned to the interface after saving. The local companion address and pairing token stay in that browser’s local storage and are sent only to the loopback companion on the same device.

No. The former public Shared gallery has been removed. Workflows, prompts, imported support files, outputs, and local-training material are private unless you export and distribute them yourself.

Yes. The Account page provides an account export and permanent deletion flow. Local companion data is stored separately on your own device and can be removed from its local data directory.

Builder 5 and Builder 10 are the defined future access tiers: Builder 5 covers Boxes 1–5 and Solo workflows; Builder 10 covers all ten boxes, Collaboration, Dual Team, and compatible GoldCap local-learning controls. Customer checkout is not active, so neither tier can currently be purchased through the site. The one configured administrator remains subscription exempt.

Yes. Visitors can explore its structure and locked states. Execution, saving, private files, provider connections, and user-owned local-model operations require sign-in and a server-side entitlement. Because customer checkout is inactive, those protected operations are currently available only to the configured administrator.

Confirm the DecaCap Local Companion is running, copy its pairing token into GoldCap, keep the default loopback address unless you deliberately changed it, and make sure the chosen model runtime is running. GoldCap’s connection check reports the exact missing layer.

Prepared pairs alone are not a training job. GoldCap also needs a base model, enough approved pairs, and a compatible fine-tuning toolkit or a command profile configured in the local companion. The job panel reports which prerequisite is missing.

Use the feedback form below for ideas, confusion, quality concerns, and feature requests. Use Support for account access, privacy, accessibility, provider failures, imports, or anything requiring a reply and follow-up.