
Multi-provider workflow forge
Build with a council
of AI models.
Mix and match primary roles, support roles, providers, and models across ten independently configurable boxes. Any entitled box can run a compatible user-owned local model; GoldCap is reserved for training documents, approved-pair preparation, and compatible fine-tuning on the user's device.

The catalog contains seven connected cloud text adapters, one user-owned local path, and seven planning-only or specialized entries. Model identifiers stay editable inside each box.
cloud text adapter wired user-owned local runtime planning only; specialized adapter required planning only; provider-dependent
The ten-box workbench
From spark to shipped,
without losing the thread.
Only the Obmil identity stays locked. Choose each box's primary role, optional support role, provider, model identifier, specialization, and route. Solo can run enabled boxes in parallel; Collaboration adds supervisor review between handoffs and an optional dual-team split.
Enter the dedicated workbench ↗









Configure every seat
Mix roles. Mix models. One board.
These are active Workbench controls—not fixed role titles or implied presets. Build the combination each project actually needs.
Primary roles
Choose a primary role independently for every available box. Reassign it whenever the project needs a different kind of specialist.
Configure the board →Support roles
Pair the primary role with an optional fact-checker, editor, tester, researcher-aide, formatter, or other secondary responsibility.
Configure the board →Providers + models
Assign providers box by box in Connections, mix several providers on one board, and edit the model identifier inside each box. The separate version note is project metadata.
Configure the board →Routing + review
Run enabled roles in parallel, order them into a chain with supervisor reviews between workers, or divide them between two reviewing supervisors in Dual Team mode.
Configure the board →
GoldCap training hub
Every box can run it.
GoldCap trains it.
Users connect their own compatible model runtime and hardware through the free loopback companion. Normal Workbench boxes use it for inference. GoldCap alone receives training documents, prepares approved Workbench pairs, starts a compatible local fine-tuning command, reports observable process state, and tests the selected runtime model.
Meet GoldCap ↗The board is open to explore
See the whole system.
Know what is active.
Visitors can inspect public pages and the Workbench's locked demonstration state. Customer checkout is not active; protected execution is currently available only to the configured administrator.