Symbiotica Client Examples

Symbiotica Client Examples solves a narrow but real LLM problem: a language model learns what a client's briefs look like by seeing several at once, and the order pipeline's other nodes hand them out one at a time. Order Assets emits records as lists, which fans the graph out and runs the lane once per asset. That's great for rendering, but for "show an LLM the shape of a brief" it means the model sees one example per run. This node collapses the set: several of the client's own briefs, numbered, as ONE string, so an LLM node downstream runs once and sees the whole population at the same time.

If you've done any few-shot prompting you know why this matters. One example teaches a pattern; five examples teach a distribution - the typical length, the vocabulary the client uses, the details they always include and the ones they skip. That's what makes a downstream prompt-writer produce briefs that sound like the client rather than briefs that merely rhyme with one.

How it works

Wire in the order, optionally narrow to a category, and it collects that type's briefs from the order sheet, drops the empty ones (a blank brief teaches nothing and reads as a missing example), and emits them as one block: a header naming the type and how many briefs follow, then each brief verbatim, numbered. The count output tells you how many made it in.

The honesty detail is the header. limit truncates - "0 keeps every asset; a smaller number keeps the first N and says so in the header." The tooltip spells out why: "a text showing three of eight examples that does not admit it reads as the whole population." An LLM that thinks three of eight is all of eight will generalize wrong; the node refuses to let that happen silently.

The inputs that matter

• order - the pipeline's order wire (required).
• category - one asset type, or empty for every type. Narrow it to the type you're writing a prompt block for.
• limit - how many briefs to include, 0–50. 0 keeps every asset of the type; a smaller number keeps the first N.

Outputs: examples (the STRING block - header + numbered briefs, verbatim from the order sheet) and count (how many briefs the text holds).

Install

ComfyUI Manager → search Symbiotica → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/symbiotica-ai/comfyui-nodes.git symbiotica
pip install -r symbiotica/requirements.txt

Restart ComfyUI. No keys or downloads - it reads the order and emits text. Zero cost.

Common issues

• "no asset carries a client brief" - the order's prompt column is empty for that type. There's nothing to show; check the order sheet.
• The header says "the first 3 of 8" and you wanted all - that's limit doing its job and admitting it. Raise or clear the limit.
• The LLM downstream keeps treating examples as literal instructions - that's a prompt-engineering issue, not a node issue: the block is data for the model, so your system prompt should frame it as "here are examples of the client's briefs," not "here is a brief to fulfill."
• Examples include blank numbered entries - they shouldn't; the node drops assets with no brief. If you still see gaps, the sheet has assets whose prompt column holds whitespace or "-".

The trap to remember: this node is about teaching an LLM the shape of a brief, not about feeding the render lane. Wire it into a prompt-author or prompt-recipe step, keep it out of the img2img path, and the numbered verbatim format does the rest.