Promptic: LLM app evaluation and prompt optimization

Promptic helps development teams measure and improve prompts and AI agents against their own examples. It combines evaluation datasets, OpenTelemetry based traces and optimization so a change can be judged on output quality, cost and response time rather than a model leaderboard alone.

What does Promptic do in plain English?

Think of it as a test bench for an AI feature. You provide realistic inputs and, where possible, expected outputs. Promptic runs candidate prompts, models or agent setups on the same cases, scores them against your chosen criteria and shows where they fail. You can then tune the setup and rerun the tests. It is for teams building AI applications; its terms exclude private consumer use, including on the Free plan.

From a dataset to a decision

Create an organization, AI Application and component, then add representative examples. For a prompt task, define the input and expected answer or structure, choose evaluators and run an experiment. Inspect individual failures before trusting an aggregate score. Compare candidates on quality, cost and latency, then keep the version that fits your actual priorities. The quickstart explains the setup; the prompt optimization guide covers the experiment loop. Good test cases matter: an incomplete dataset cannot prove that every production request will work.

Traces and agent optimization

Tracing records LLM calls, tool calls and steps with token, cost and timing information per span. It can turn a failed output into a concrete debugging path. For agents, Promptic also compares whole execution variants on the same cases, including multi step behavior and tool use. The agent optimization guide describes that workflow. An example diagram or benchmark on the marketing page is illustrative; use results from your own workload.

Plans, model charges and eligibility

As checked on 27 September 2026, Free is €0 for one user and one AI Application, with 200,000 Promptic Credits per month and 14 days of trace retention. Team is listed at €149 per month and Business at €599 per month, with higher allowances and governance features. Bringing your own model keys is available; managed model usage can be billed separately. Check the current pricing and plan limits before deciding. The Free label does not remove the business use restriction.

Data to review before connecting production

Traces and evaluations can contain prompts, model inputs and outputs, tool arguments and metadata. The terms say customers retain ownership of their content and it is not used to train foundation models without a separate written agreement. They also restrict highly sensitive submissions without written agreement. Model provider and inference routes can have different regions and retention; review the exact route before sending real data. Start with sanitized test cases and read the terms and provider conditions.

Common questions

Can I evaluate an AI agent as well as a prompt?

Yes. Promptic describes prompt optimization and agent variant evaluation. Use the same cases and metrics for a meaningful comparison; inspect traces for failures.

Is there a free plan?

Yes, with one user, one AI Application and monthly credit and retention limits. It is for eligible business or professional users, and model charges may be separate.

Does Promptic guarantee EU only processing?

No blanket guarantee follows from an EU platform data region. Inference location and retention depend on the selected model route, so check the provider configuration and contract.

Official resources

• Product documentation for setup, SDK and workflows.
• Current pricing for limits and extra charges.
• Company contact page for account or enterprise questions.