DIGIT Console Doctor

Every ComfyUI user knows the moment: the graph throws, the console fills with a wall of traceback, and you're squinting at ModuleNotFoundError trying to remember what that was. The DIGIT Console Doctor automates the squinting. It reads ComfyUI's in-memory log buffer, filters it down to the errors and warnings that matter, and sends them to Gemini for a plain-language diagnosis with fix suggestions.

It's the pack's own support node, and it's genuinely handy for the pack's own nodes - the DIGIT family has enough cloud calls and env-var dependencies that when something breaks, it's usually an auth or configuration error with a recognizable signature. This node turns "figure out what this traceback means" into "read what Gemini said it means."

How it works

Under the hood it pulls entries from ComfyUI's built-in log buffer (app.logger), filters by severity, and sends the top slice to a Gemini model with a system prompt that instructs it to act as a ComfyUI debugger: identify the root cause of a Python traceback, suggest the pip install for a missing module, offer VRAM/driver fixes for CUDA errors, and name the node that failed. The system prompt is built in - you don't have to write it.

The inputs that matter:

• trigger - a plain INT you bump to force a re-run. The node needs a way to know you want a fresh diagnosis after you've changed something, so this is the "run it again" button.
• log_level - errors_only, warnings_and_errors (default), or all. Start with errors-only when it's noisy.
• max_log_entries - how many log lines get sent (default 50). Enough for a traceback, not enough to burn tokens.
• node_filter - show only logs containing a node name, so you can focus on one suspect.
• extra_context - tell it what you were doing ("I was loading a LoRA"), which meaningfully improves the diagnosis.
• run_after - connect any output from your last node here to force Console Doctor to execute last.

Outputs are diagnosis (the plain-language explanation with fixes) and raw_logs (what was actually sent, so you can double-check it didn't miss anything).

Installing it

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/thedepartmentofexternalservices/comfyui-digit.git
cd comfyui-digit
pip install -r requirements.txt

Or ComfyUI Manager → search comfyui-digit → install → restart. This one does need the GCP setup because the diagnosis runs on Gemini through Vertex AI: gcloud auth application-default login, a Vertex AI–enabled project, and gcp_project_id set or auto-detected.

Where it trips people up

The trigger is the part people miss - if you fix something and re-run the graph, the node may not re-fire unless the trigger value changed, so bump it. Also be aware of the scope: it reads the in-memory log buffer of the running process, so it sees recent activity in this ComfyUI session, not your entire log history. If the buffer was flushed or the error happened at startup before the node existed, there may be nothing to diagnose. And like every cloud call, the log content leaves your machine - the node sends a filtered slice to Google, so don't run it with log_level: all on a log full of API keys you don't want shipped anywhere.