DecaCapBuilders
AccountOpen workbench

Platform map

A workshop with rules,
not a wall of chat boxes.

DecaCap keeps provider choice, specialist role, handoff order, entitlement, local safety, and final review visible from beginning to end.

Explore the board →Compare plans
01

Model windows

  • Provider + editable model identifier
  • Primary and support roles
  • Light / medium / heavy response-effort guidance
  • Ready, running, streaming, failed, disabled, and locked states
  • Chat and read-only code views
  • Solo user stop, Queen-supervisor stop/takeover, retry, handoff, expand, copy, disable, prompt, and clear controls
  • Streaming cloud-text output; completed-response local output
  • No fixed worker-level execution cutoff
02

Orchestration

  • Solo parallel execution across enabled entitled boxes
  • Sequential Collaboration handoffs
  • Box 10 Queen supervisor in Collaboration mode
  • Box 5 second Queen supervisor in Dual Team mode
  • Dual 1–4 and 6–9 team split
  • Queen-owned timed checks against explicit stop/takeover rules
  • Supervisor review appended between worker handoffs
03

Project control

  • Per-box task and specialization assignments
  • Separate supervisor review rules and project outlines
  • Board or supervisor messaging
  • @provider routing by provider ID
  • Machine-readable PDF and DOCX text extraction
  • Incremental approved-pair preparation
  • Markdown, text, Word, PDF, and JSON exports; unchanged GLB pass-through only after a real GLB import
04

User-owned local model

  • Per-user loopback companion instead of a central bridge
  • The same local text model can be selected in any entitled Workbench box
  • GoldCap is the only training-document and training-control destination
  • Explicit opt-in capture of completed prompt/output pairs
  • Automatic local batching while the Workbench page remains open
  • MLX or user-configured compatible fine-tuning command
  • Observable process state, log-derived progress when available, pause/resume where the operating system supports it, stop, and output path
  • Testing of the currently selected runtime model without a provider API key

Structural identity

Ten boxes.
Canon locked.

Provider, model identifier, task, routing position, and workflow roles stay configurable; the separate version field is a project note only. The Obmil identity never follows those role changes: Box 1 remains Planar Obmil, Box 5 becomes Majesty Obmil only in Collaboration Dual Team mode, Box 6 remains Sunshine Obmil, and Box 10 is Cleaner Obmil only in Solo mode and Majesty Obmil in every non-Solo mode.

Planar Obmil avatar for Box 1BOX 1Planar Obmil
Curious Obmil avatar for Box 2BOX 2Curious Obmil
Raging Obmil avatar for Box 3BOX 3Raging Obmil
Aqua Obmil avatar for Box 4BOX 4Aqua Obmil
Wicked Obmil avatar for Box 5BOX 5Wicked Obmil → Majesty Obmil
Sunshine Obmil avatar for Box 6BOX 6Sunshine Obmil
Urkel Obmil avatar for Box 7BOX 7Urkel Obmil
Greedy Obmil avatar for Box 8BOX 8Greedy Obmil
Ivy Obmil avatar for Box 9BOX 9Ivy Obmil
Cleaner Obmil avatar for Box 10BOX 10Cleaner Obmil → Majesty Obmil

Provider catalog

Fifteen entries, four honest states.

Seven entries have wired cloud text-output adapters, one connects to the user's local text runtime, and seven remain planning-only until their specialized or official-compatible adapters exist. A listed default is editable and does not guarantee that a user's provider account can access that model.

OAOpenAI
cloud text wired
gpt-5.6-terra

Text-output adapter available when a valid user key and model identifier are supplied. This Workbench does not send image inputs.

CLClaude
cloud text wired
claude-sonnet-5

Anthropic Messages text-output adapter available when a valid user key and accessible model identifier are supplied.

KMKimi
cloud text wired
kimi-k3

Kimi OpenAI-compatible text-output adapter available when a valid user key and accessible model identifier are supplied.

GMGemini
cloud text wired
gemini-2.5-pro

Google Generative Language text-output adapter available when a valid user key and accessible model identifier are supplied.

NONotion AI
planning · provider dependent
provider-managed

Notion does not expose a general Notion AI model endpoint for this workflow; execution remains unavailable until a compatible official API exists.

PXPerplexity
cloud text wired
sonar-pro

OpenAI-compatible text-output adapter available when a valid user key and accessible Sonar model identifier are supplied.

ELElevenLabs
planning · adapter needed
eleven_multilingual_v2

Selectable for workflow planning; media execution requires its specialized output adapter.

FFAdobe Firefly
planning · provider dependent
firefly-image

Requires an eligible Adobe developer integration and provider approval.

DSDeepSeek
cloud text wired
deepseek-v4-flash

OpenAI-compatible text-output adapter available when a valid user key and accessible model identifier are supplied.

CRCreao AI
planning · provider dependent
provider-managed

No verified general model endpoint is wired; shown for catalog compatibility only.

GRGrok
cloud text wired
grok-4.5

xAI-compatible text-output adapter available when a valid user key and accessible model identifier are supplied.

SISider AI
planning · provider dependent
provider-managed

No verified general provider endpoint is wired; shown for catalog compatibility only.

KLKling
planning · adapter needed
kling-video

Selectable for workflow planning; generation requires Kling account/API eligibility and a specialized media adapter.

MYMeshy
planning · adapter needed
meshy-4

Selectable for 3D workflows; generation and real GLB retrieval require the Meshy task adapter.

LMLocal Model
user-owned local
use-paired-device-model

Any entitled box can run the user’s paired loopback model. GoldCap alone handles training documents and compatible training controls. No site-owned model, GPU, endpoint, or API key is required.

GoldCap training-hub characterUSER-OWNED LOCAL TRAINING

GoldCap training hub

Your machine.
Your model. Observable state.

Pair the free device-local companion and select that compatible local text model in any entitled Workbench box. GoldCap is the single destination for training documents, approved completed pairs, preparation, and compatible local fine-tuning controls—without sending model files or adapters to DecaCap.

01

Training interface

Training-document import, pair counts, local process state, log-derived progress when exposed, errors, supported lifecycle controls, and runtime-model testing.

02

Device settings

Loopback address, pairing token, Workbench runtime model, base model, toolkit profile, data scope, format, and automatic batch threshold.

Download Local Companion