Quick Start¶

This page has one job: help you complete one real workflow with the shortest possible path.

Warning

Prompt Optimizer does not include a ready-to-use model by default. Before your first run, configure at least one text model in Model Management, or analysis, optimization, testing, and evaluation will not work.

First-time users: only do these 4 steps¶

Configure one text model in Model Management
Open the workspace that matches your input structure
Run one Optimize or Analyze action on the left
Run Test / Result Evaluation / Compare Evaluation on the right

If you are not sure where to go next:

model setup: see Model Management
workspace choice: see Choose Workspace
analysis vs evaluation: see Testing & Evaluation

Step 1: configure the minimum model setup¶

text workspaces: at least 1 text model
image workspaces: at least 1 text model + 1 image model

For most first-time users, the simplest path is:

configure one text model
run left-side analysis or optimization
run right-side testing
run one evaluation

Once that works, add a second text model only if you actually want side-by-side comparison.

See Model Management and Model Testing Strategy for more detail.

Step 2: choose the right workspace¶

Workspace	What you optimize	Main right-side input
System Prompt Workspace	roles, rules, boundaries, output policy	one test message
User Prompt Workspace	one direct task prompt	usually no extra input
Variable Workspace	a reusable prompt template	one set of variable values
Context Workspace	one target message inside a conversation	full conversation + shared variables + optional tools
Text-to-Image Workspace	image prompt	image model
Image-to-Image Workspace	image-to-image prompt	input image + image model

Step 3: understand left vs right¶

This is the most important product boundary in the app:

left side edits prompts
right side runs real outputs
right-side evaluation judges real execution evidence

For the full boundary, see Testing & Evaluation.

Step 4: run one minimum workflow¶

In any text workspace, you can start with this sequence:

enter the original content on the left
click Optimize or Analyze on the left
move the result into the lower workspace area
choose original / workspace / vN on the right
run one or more tests
start with Result Evaluation
then use Compare Evaluation if you have multiple columns
apply valuable suggestions back to the left workspace

If model connection fails, then check deployment and connection environment¶

Your situation	Recommended path	Why
public HTTPS API only	Web / hosted version	simplest setup
Ollama, LM Studio, local network, internal API	Desktop app	avoids common browser CORS and mixed-content limits
self-hosted page plus MCP	Docker	packages the web UI and MCP together

Warning

The web version is not a built-in proxy layer. The browser still sends requests directly to the model service you configure.

Three small examples¶

Example A: System Prompt Workspace¶

Left side:

You are a technical support assistant. Answer user questions clearly.

Right-side test message:

My app crashes on Windows at launch. What should I check first?

Example B: User Prompt Workspace¶

Left side:

Write a poem about autumn.

No extra test text is needed on the right in this workspace. The user prompt itself is the tested object.

Example C: Text-to-Image Workspace¶

Left side:

An orange cat standing on a neon street in the rain, cinematic mood, rich detail, low saturation.

Then:

choose a text model for left-side prompt work
choose an image model on the right
compare prompt versions through real generated images

Common mistakes¶

“Web deployment” means there is a proxy layer¶

No. The browser still connects directly to your configured model service.

Why can’t I connect to my local model?¶

The most common reason is browser-side CORS or mixed-content restrictions. In those cases, the desktop app is usually a better fit.

Will clearing browser data erase my setup?¶

Yes. Browser-side data is stored locally in the browser. Export from data management before clearing it.