Four steps to your first useful workflow
You do not need to enable every optional component on day one. Confirm the core chat path first, then expand intentionally.
Launch and initialize
Run MRCO AI. Setup may check or download runtime files and the default local model if they are missing.
Configure a model
Open Models/Online, set a compatible endpoint and model, add a key if needed, then use Test or Fetch.
Try a simple prompt
Ask the agent to introduce itself, then try a low-risk task such as opening Notepad.
Add optional features
Install voice, vision, browser testing or embeddings only when the workflow needs them.
Make the model path predictable first.
MRCO AI supports configuration for compatible endpoints and model IDs. Provider names and model support can change, so test the exact model you plan to use.
- ✓Confirm the endpoint and final chat-completions route.
- ✓Enter and protect the API key if the provider requires one.
- ✓Fetch the model list or type the exact model name.
- ✓Test connectivity and tool calling separately.
- ✓Choose Local, Online or Hybrid based on your constraints.
[x] App opens successfully
[x] Provider endpoint validated
[x] Model returns a basic response
[ ] Tool Calling tested
[ ] Read-only Windows action tested
[ ] Optional features installed as needed
LOG FILE
%LOCALAPPDATA%\AiDesktop\crash.log
Know what is required—and what is optional
These answers reflect the supplied user guide and the current documented implementation.
Does MRCO AI run entirely offline?
Not necessarily. Local text or vision inference can run on the machine, but web search, page retrieval, downloads, online providers and the Edge TTS fallback can use the network. Choose the local mode and avoid online tools when a task must remain offline, then verify the full workflow's data path.
Can I use any online model?
Use a provider compatible with the expected Chat Completions flow and test it. Tool calling is model-dependent; a model that can answer a message may still produce weak or failed tool calls.
Does Hybrid always produce a better answer?
No. It can coordinate eligible providers and combine outputs for supported tasks, but it may increase latency and API cost and is not always more accurate.
Can it perform real cryptocurrency trades automatically?
The guide describes market-data analysis, internal positions, scanning proposals and backtesting. It explicitly notes that internal close actions do not themselves send real exchange orders. Any real execution integration requires separate verification of connectivity, permissions, risk controls and logs.
Where are conversations and settings stored?
The documented data root is %LOCALAPPDATA%\AiDesktop. It can contain settings, provider configuration, the local database, models, runtimes, skills, custom extensions and a separate browser profile. Protect this folder; the provider JSON can contain unencrypted API keys in the current code path.
What if the model answers but does not use tools?
Test Tool Calling in the model window, use an instruction-following model with tool-call support and verify the relevant tools are available in the current request. When native Tool Calling fails, a prompt-based fallback may be less reliable.
Can MRCO AI build and test software?
Yes—MRCO AI can create applications and test them. It can generate project files, use installed development tools, run build and test commands, inspect output and logs, diagnose errors, attempt fixes and rerun checks. If a required supported tool is missing, it can check for and install tools when the environment and permissions allow. The final outcome still depends on the selected model, toolchain, test coverage and project requirements.