Promptic

Promptic, from Promptic GmbH of Frankfurt am Main, is a platform that tunes applications built on large language models. It helps a team compare candidate prompts, models, agent designs and tool choices against its own examples, so the setup it ships is the one that scores highest.

What Promptic offers

• Benchmarking: scores each candidate on quality, cost and latency together, using the same task and examples.
• Prompt and model tuning: optimises prompts and model choices against the team's own data and measures.
• Agent and tool tuning: covers agent design and tool choice, including tools found on an MCP server.
• Extraction scoring: checks structured output field by field.
• Tracing: records each model call, tool call and step through OpenTelemetry, with a Python SDK.

Customers and markets

Teams building applications on large language models. Under its terms, use is limited to business customers, and sensitive personal categories are excluded unless both sides agree in writing.

Key facts

• Based in: Frankfurt am Main
• Data location: held in the EU
• Opened to all: 25 September 2026, after running by invitation before that
• Held-out data: holds back a test split, so tuning is never scored on the examples it learned from