Quick answer: If you are iterating on LLM prompts alone or in a pair, free tiers (Langfuse, Helicone OSS, PromptLayer free) cover it. Mid-size teams with production prompts need one observability platform plus one evaluation platform. Enterprise with regulated-industry constraints should buy Humanloop or W&B Prompts. Do not buy anything until you have ten prompts in production.

The four jobs-to-be-done

Versioning. Track every prompt change with author, diff, and rollback.
Observability. Log every completion with inputs, outputs, model, cost, latency.
Evaluation. Run test suites against new prompt versions before deploy.
Collaboration. Non-engineers edit prompts safely with review and approval.

Vendor breakdown

Langfuse — free OSS + £29-£199/mo cloud

Best at observability. Strong tracing, cost dashboards, self-host option. Thin on evaluation workflows.

Helicone — free OSS + £20-£200/mo

Similar space. Lightweight gateway-style integration. Good free tier. Fewer eval features than PromptLayer.

PromptLayer — £20-£199/mo

Strongest at versioning + prompt marketplace. Non-engineer-friendly editor. Integrations with OpenAI, Anthropic, Mistral.

Humanloop — enterprise

Best at evaluation and regulated-industry workflows. Human-in-the-loop feedback. Typical ARR £10,000-£40,000+.

Weights & Biases Prompts — enterprise

Excellent if you already use W&B for ML. Ties prompt experiments to ML experiments. Not worth adopting just for prompts.

Built-in platforms (OpenAI Playground, Anthropic Console)

Free. Fine for solo exploration. No real versioning or collaboration.

Decision framework

<10 prompts, 1-3 engineers: no tool. Git + a .md file. Graduate later.
10-50 prompts, team of 5-15: Langfuse OSS or PromptLayer standard.
50+ prompts, PM/content teams editing: PromptLayer Pro + evaluation harness.
Regulated industry: Humanloop or W&B.

Prompt-Engineering Tool Buyer's Guide (2026)