ComfyDeploy API Cache Run IDs (Soze)

ComfyDeploy - a separate hosted platform for running ComfyUI workflows as an API - hands you back a run_id when you kick off a job. Fine for one run; annoying the moment you're queueing a batch of ten and want to come back later and check on all of them. A single ComfyUI graph execution has no memory of what happened on a previous run, so if you want that list of IDs to survive across separate queue executions, you need somewhere to keep it. This node appends one run_id to a local cache file every time it fires.

How it works

You give it run_id (the ID you just got back from a run) and cache_save_folder (where the cache file lives), and it appends that ID into the cache. There's no output - this is a pure side-effect node, purely about the write.

It's built to be one of a matched trio: ComfyDeploy API Cached Run Info reads the accumulated list back, and ComfyDeploy API Clear Cached Run IDs empties it once you're done with a batch. All three key off the same cache_save_folder - that shared path is what ties them together into one cache, not any direct wiring between the nodes themselves.

The inputs and outputs that matter

• run_id (required, STRING) - the ID to append.
• cache_save_folder (required, STRING) - where the cache file lives.

No outputs - flagged as an output node, a dead end in the graph by design.

How to install it

ComfyUI Manager → search "ComfyUI_Soze" → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/SozeInc/ComfyUI_Soze.git
pip install -r ComfyUI_Soze/requirements.txt

then restart. This node writes to a local file only - no network call, so it doesn't need CD_API_KEY itself. That key is needed elsewhere in your setup, for whatever actually triggers the ComfyDeploy run and produces the run_id you're feeding in here.

Common issues & troubleshooting

Because three separate nodes all key off cache_save_folder, a typo or a trailing-slash mismatch between them means this node writes to one file while Cached Run Info reads from what it thinks is the same path but isn't - and you'll come up empty for no obvious reason. Hardcode the exact same string into all three (or wire it from a single Output Filename node so there's only one place to edit) rather than retyping the path in each.

If you're not seeing IDs accumulate across a batch, double-check this node is actually firing on every queue execution and not being skipped by some upstream gate - since it has no output to confirm success with, the only way to verify it worked is checking Cached Run Info's count afterward.