'/%3e%3cpath%20d='M18%2077C18%2050%2036%2031%2062%2031C74%2031%2084%2035%2092%2042L79%2057C74%2053%2069%2051%2062%2051C50%2051%2041%2061%2041%2077C41%2092%2050%20102%2063%20102C70%20102%2076%2099%2082%2094L95%20108C87%20117%2076%20123%2062%20123C36%20123%2018%20104%2018%2077Z'%20fill='url(%23b)'%20opacity='.72'/%3e%3cdefs%3e%3clinearGradient%20id='a'%20x1='33'%20y1='26'%20x2='112'%20y2='118'%20gradientUnits='userSpaceOnUse'%3e%3cstop%20stop-color='%2346F0FF'/%3e%3cstop%20offset='1'%20stop-color='%23FF4FD8'/%3e%3c/linearGradient%3e%3clinearGradient%20id='b'%20x1='18'%20y1='31'%20x2='95'%20y2='123'%20gradientUnits='userSpaceOnUse'%3e%3cstop%20stop-color='%238CFFC1'/%3e%3cstop%20offset='1'%20stop-color='%238F62FF'/%3e%3c/linearGradient%3e%3c/defs%3e%3c/svg%3e)
Local-first by default
Run Ollama, llama.cpp, and local models on your hardware before ever touching a paid API. Escalate only when the job genuinely needs it — not as the only path forward.
Local-first AI harness • agent cockpit • workflow engine
Run AI your way.
Keep the keys.
A local command center for models, agents, and workflows — designed to run on your hardware by default and reach for APIs only when the job calls for it. Yours to control, yours to build on.
cadela boot --mode local-first
> loading agents: code, media, docs, review
> model router: ollama + deepseek + api fallback
> workflows: comfyui, ue5, prompts, files
> clean-room docs: enabled
> status: ready to conquerWhat Cadela is
The harness between you and the models.
Cadela is being built as the connective layer between local LLMs, API models, ComfyUI pipelines, Unreal Engine tooling, code agents, documentation workflows, and project-specific skills — one workspace that puts you in control instead of locking you into a cloud dashboard.
The philosophy
Why local-first puts you back in control.
Most AI platforms route everything through their cloud — your context, your files, your prompts — as the only option. Cadela flips that. Local models run on your workstation with zero data leakage and no per-token cost for daily work.
External APIs become a deliberate choice rather than the default. You decide what's sensitive enough to keep local and what's worth sending upstream. Your workflow, your hardware, your call.
Private by default. Powerful when you need it.
Multi-agent workflows
Dedicated roles for planning, coding, reviewing, clean-room documentation, workflow generation, prompt building, and media direction — coordinated from one harness instead of a dozen tabs.
Built for real projects
Game dev pipelines with UE5, AI media production through ComfyUI, clean-room source analysis, community bots, and prompt cookbooks — the harness bends to the project, not the other way.
Model routing
One frontend. Many brains.
Cadela's long-term goal is to route work between local models and external APIs based on cost, privacy, speed, context size, and task type — automatically or by explicit choice.
Dive into the architecture →Local LLMsOllama, llama.cpp, Qwen, DeepSeek-style code models — run on your hardware.
Creative pipelinesComfyUI integration for image, video, audio, LoRA training, and prompt builders.
Game dev hooksUE5 tooling, C++ guidance, Blueprint-safe architecture planning, and procedural helpers.
Project memoryReusable skills, project rules, prompt packs, and system docs stored per repository.
Get involved
Built in public. Yours to join.
Cadela is being developed openly. Follow the build log, join the conversation, or contribute.