Using the Platform
Run prompt optimization from the Future AGI UI: pick a dataset and column, configure prompt and evals, run optimization, and apply the best prompt.
What it is
Using the platform for optimization means running prompt optimization from the Future AGI web UI instead of code. You open a dataset, click Optimize to open the Run Optimization panel, then set the run name, the column that holds the prompt template, the optimizer (e.g. GEPA), the language model, and optimizer parameters (e.g. Max Metric Calls), and add evaluations. You click Start Optimization and the run executes on Future AGI’s backend; when it finishes, you review results in the Optimization tab, compare scores across variations, and apply the best prompt. No Python or SDK required—everything is driven by the UI and your existing datasets and evals.
Use cases
- No-code workflow — Improve prompts without writing code; use the UI for configuration and runs.
- Dataset-centric — Optimize a prompt that already lives in a dataset column; data and results stay in the platform.
- Team visibility — Runs and results are stored in Future AGI so others can see and reuse them.
- Reuse existing evals — Pick from preset or previously configured evaluations instead of defining them in code.
- Iterative refinement — Run optimization, apply the best prompt, then run again if you want to refine further.
How to
Open the optimization panel
Go to the Dataset view and open a dataset that has the inputs and (if needed) model-generated outputs you use for optimization. In the top action bar, click Optimize (next to Run Prompt, Experiment, and Evaluate). Choose the dataset column that contains the prompt you want to improve.
Set general details
In the Run Optimization panel, fill in:
- Name — Give the run a clear name (e.g. GEPA-Feb27-1655) so you can find it later in the Optimization tab.
- Choose Column — Select the dataset column that contains the prompt template to optimize. The prompt column is used as the baseline for the run.
- Choose Optimizer — Pick an optimizer (e.g. GEPA, Bayesian Search, Meta-Prompt, ProTeGi, Random Search, PromptWizard). Each has different trade-offs between speed and quality.
- Language Model — Select the model used for optimization (inference and/or teacher model, depending on the optimizer).
Add parameters
In the Add Parameters section, set optimizer-specific options. For example, Max Metric Calls limits the maximum number of metric evaluations; the suggested value is tuned for a good balance between speed and quality. Parameters vary by optimizer (e.g. num_rounds, beam_size). Use the recommended defaults unless you need to tune them.
Add evaluations
Open the Evaluations section and select the evals to run on your dataset. Add and configure the evaluation metrics that will score each prompt variation. The optimizer uses these scores to rank variations and pick the best prompt.
Start the optimization
When Name, Choose Column, Choose Optimizer, Language Model, parameters, and evaluations are set, click Start Optimization. The run executes on Future AGI’s backend; progress and results appear in the Optimization tab. Use Cancel to close without starting.
Review results
After the run completes, open the Optimization tab for that dataset or run:
- Compare variations — The system shows multiple optimized prompt versions ranked by evaluation scores.
- Check scores — A table lists each prompt with its scores (e.g. Context Relevance, Context Similarity); the original prompt’s score is included for comparison.
- Pick the best — Review the top variations; the best-performing prompt is highlighted. You can inspect each one before deciding.
Apply the optimized prompt
When you’ve chosen the best version:
- Apply the optimized prompt so it replaces the original in your dataset or workflow.
- Export the updated dataset if you need it elsewhere.
- Run another optimization if you want to iterate further.
What you can do next
Optimize your first prompt
Run optimization from code with the agent-opt library and a small dataset.
Dataset optimization
Deep dive on dataset optimization: concepts, optimizers, and best practices.
Using the Python SDK
Full control over optimizers and evaluation in your own code.
Optimization overview
Platform vs SDK and when to use which.