Promptmetheus
Promptmetheus is a prompt engineering IDE for teams building LLM apps, agents and workflows. Compose prompts from reusable blocks, test them across 150+ models from 15 providers, then compare ratings, inference costs and full version history.
What is Promptmetheus?
Promptmetheus calls itself a Prompt Engineering IDE: the same idea as VS Code or PyCharm, but pointed at prompt design rather than code. Instead of typing a prompt into a box and hoping, you assemble it from LEGO-like blocks — Context, Task, Instructions, Samples (shots) and Primer — and each section can hold several variations that you combine and compare. The stated aim is minimum cost and maximum performance.
Twelve named features carry that promise: prompt composition, project- or prompt-level variables, custom evaluators that check each completion against your constraints, a model catalogue, projects grouping prompts and completions, test datasets that inject dynamic context to simulate user data or retrieved content, completion ratings, cost calculation, full traceability through versioning and changelogs, stats and insights, real-time sync, and data export to .txt, .csv, .xlsx or .json.
The catalogue spans 15 API providers and more than 150 models — Anthropic, OpenAI, Google DeepMind, Mistral, Perplexity, xAI, DeepSeek, Cohere, Groq, Fetch.ai, OpenRouter, AI21 Labs, Venice, Moonshot AI and Deep Infra — and any model exposed through an OpenAI-compatible API or the LiteLLM SDK can be added.
Two separate, unconnected products exist. Forge is free, runs entirely offline with data kept in the browser, and needs no account; it is in maintenance mode, so new features land only in Archery, the paid cloud IDE with sync, traceability and real-time collaboration.
The scope is stated plainly. Promptmetheus designs, tests and optimizes individual prompts; it does not assemble or run agents. It presents itself as complementary to LangChain and LangFlow, which deploy the agents while Promptmetheus tunes each prompt in the chain, and as a step beyond provider playgrounds, which offer neither traceability nor experimentation tooling. The publisher also advances his own notion of an AIPI, an AI Programming Interface: endpoints that mediate LLM interactions through hosted prompts rather than static code.
Behind the product stands one individual, Toni Engelhardt, with residence and jurisdiction in Portugal. It is privately owned, described as profitable, and not seeking venture funding. A 12-inch screen is the hardware minimum. Side resources include an LLM Index, an LLM Knowledge Base, LLM Benchmarks, a blog, documentation and a Discord community.
What it does
- Compose a prompt block by block — context, task, instructions, samples and primer — then iterate on variations of each block.
- Test the very same prompt across more than 150 LLMs from 15 providers.
- Evaluate every completion automatically against constraints you define, and rate its quality by hand.
- Compare results per model and per variant through built-in statistics.
- Estimate inference cost before anything is deployed.
- Trace every change to a prompt with detailed versioning and changelogs.
- Collaborate in real time on a shared workspace, and export prompts and completions to .txt, .csv, .xlsx or .json.
When to use Promptmetheus / When not to
A quick filter to help you decide if Promptmetheus is the right fit.
When to use Promptmetheus
- AI and product teams shipping LLM-powered features, who need prompt design to become a documented, repeatable process instead of trial and error.
- Engineers who must compare the same prompt across several providers and models before committing, and who want inference cost estimated before anything reaches production.
- Builders of agent chains and multi-step workflows, where each prompt has to be tuned on its own because errors accumulate along the chain.
- Distributed teams on the Team plan, working together in real time on a shared prompt library with user management and per-user private workspaces.
- Lecturers, researchers and AI consultants sharing prompt work with students or clients: separate projects, full traceability, and announced discounts for educational use. Berkeley, UMD and UNSW appear as university users on the partnerships page.
When not to use Promptmetheus
- Teams looking for a platform that also assembles and runs agents: Promptmetheus covers only the design, testing and optimization of individual prompts.
- Developers who need programmatic access. There is no API and no SDK today; both sit on the roadmap.
- Anyone relying on no-code automation platforms: there is no direct integration with Make, Zapier, IFTTT or n8n, so prompts have to be copied over by hand.
- Mobile and tablet users: a 12-inch screen is the stated minimum, and no iOS or Android app exists.
- Users expecting an all-inclusive subscription: inference is billed to your own provider API keys on top of the plan, refunds are issued only for an incorrect charge, and the free offline Forge edition is in maintenance mode, so new features land only in the paid cloud IDE.
How to use Promptmetheus
A typical end-to-end flow, from setup to results.
- Open Forge in your browser to try the workflow immediately: no installation, no account, everything stored locally and working offline.
- For the cloud IDE, register an account on Archery with an email address and a password; every paid plan opens with a 7-day free trial.
- Add your own API keys from the LLM providers you intend to use — stored locally on your device by default, synced optionally depending on the plan.
- Create a project to group your prompts, datasets and completions in one place.
- Compose the prompt block by block, then create variations for each block.
- Define variables at project or prompt level for the details that keep recurring.
- Load a test dataset to inject dynamic inputs and simulate real user data or retrieved content.
- Run the prompt on one or several models, then rate the completions you get back.
- Read the dashboard: statistics per model and per variant, plus the estimated inference cost.
- On the Team plan, open a shared workspace, manage users and work together in real time; product documentation and a troubleshooting section live on docs.promptmetheus.com.
Pros & Cons
Pros
- Very broad model coverage: 15 providers, more than 150 LLMs, plus any custom model reachable through an OpenAI-compatible API or LiteLLM.
- Full traceability of the design process through versioning and changelogs — precisely what provider playgrounds do not keep.
- Automatic evaluators and completion ratings make variants comparable on figures rather than impressions.
- Inference cost estimated before deployment, so the production bill is known in advance.
- A permanently free tier, Forge, that needs no account and works offline, plus a 7-day free trial on every paid plan and announced discounts for educational use.
- Real-time collaboration and a shared prompt library on the Team plan; API keys stored locally on the device by default; data exportable in four formats.
- Unusually legible legal and security posture: subprocessors named one by one, a named and reachable DPO, GDPR compliance claimed explicitly, a detailed security page with a security report on request, and a public status page carrying 90 days of uptime history.
Cons
- No API and no SDK: programmatic integration is impossible today, and no native connector exists for automation platforms such as Make, Zapier, IFTTT or n8n — prompts have to be copied across by hand.
- Inference is not included: provider API keys, and the bill that comes with them, sit on top of the subscription.
- A 12-inch screen minimum rules out phones and compact tablets, and there is no iOS or Android application.
- Forge is in maintenance mode: the free offline edition receives no new features.
- Restrictive refund policy — nothing outside an incorrect charge, and no refund for time already paid but not consumed.
- No certification of its own: SOC 2 and ISO are announced as in progress and only PCI compliance is claimed; no DPA is published or offered; no hosting country or region is disclosed; no retention period is quantified; and no opt-out from model training is documented.
- The publisher is a single individual with no published legal entity or postal address, the interface exists in English only, and the Team plan has no self-serve price beyond three seats (19 USD per month per extra seat, Enterprise on request).
Pricing & Plans
Promptmetheus offers a permanently free plan, Playground, which runs the Forge editor for a single user with local data storage, OpenAI models and community support. The lowest paid entry point is the Single plan at 29 USD per month for one user, followed by Team at 99 USD per month covering three users, then 19 USD per month for each additional user; Enterprise pricing is quoted on request and not published. Billing runs monthly or annually, is processed through Stripe, and every paid plan opens with a 7-day free trial. Inference is not included in any plan: users must supply their own provider API keys and are billed separately by those providers for consumption. Refunds are not available except where a charge was made in error, and cancellation takes effect at the end of the current billing cycle.
- the Forge editor
- 1 user
- local data storage
- OpenAI models
- Stats & Insights
- data import and export
- community support.
- the full Prompt IDE for 1 user
- cloud sync across devices
- 15 providers and 150+ models
- multiple projects
- automatic evaluators
- prompt history and full traceability
- Stats & Insights
- data export
- everything in Single for 3 users included
- then 19 USD per month per additional user
- plus user management
- a shared workspace with real-time collaboration
- and business support.
- Enterprise — no published price
- quoted through the sales contact.
Data, GDPR & hosting
A consolidated view of how Promptmetheus handles your data.
GDPR overview
GDPR compliance is claimed explicitly: the privacy policy states it was designed to comply with the General Data Protection Regulation. A data protection officer is named — Toni Engelhardt, reachable at toni@promptmetheus.com. The operator is established in Portugal, inside the European Union, so no Article 27 representative applies. Access requests go by email to contact@promptmetheus.com and are compiled by the publisher; erasure can be triggered by deleting the account inside the application, or requested by email, with one exception: database backups held on the publisher's servers and backup devices are exempt, and shared with no one. Ten subprocessors are named individually: Gandi, Digital Ocean, Vercel, Notion, Stripe, Sentry, Plausible, PostHog, Microsoft Clarity and Rewardful — Plausible presented as GDPR-compliant and free of tracking cookies. The policy took effect on 9 October 2025, with a dated changelog running back to 15 June 2023. Minimum age is 16 without guardian consent.
Who owns the data?
Content you add — text, links, files, images, video — remains yours. The terms make you fully responsible and liable for it, allow you to upload only what you own or hold explicit permission for, and have you grant the publisher permission to store that content on his servers. No transfer of ownership and no wider licence are claimed over it. The publisher separately declares himself owner or licensee of the intellectual property of the site itself. Your prompts, datasets and embeddings sit in the platform database, where staff may access them for maintenance, improvement, usage analysis and support. Third-party API keys default to local storage on your own device, with database sync optional depending on the plan.
Reuse rights
The declared uses are narrow and written down. Your email address serves communication and is not shared without explicit consent, save where the law compels it. Usage data is compiled into general statistics used to improve the service. General data — prompts, datasets, embeddings — is not shared with third parties, with two stated exceptions: the named subprocessors, and the AI providers themselves. That second exception matters in daily use: running a prompt sends the prompt and its related data to the provider you selected, which may store and process it, along with metadata about the request, under its own policy rather than this one. Personal data collected covers email address, a password hash following NIST guidance, profile details, correspondence and device information. Session replays through Sentry and Microsoft Clarity run with sensitive data redacted, and no tracking scripts load outside the signed-in area. No document anywhere mentions training models on customer data.
Data retention & training
Hosting summary
Logs and everything you submit are stored on Digital Ocean and Vercel cloud servers. Digital Ocean carries the database and backend services, which is where the bulk of personal and general data lives; Vercel handles DNS, hosting for all frontends, request and error logs, and part of authentication, and itself relies on AWS. No hosting country or region is disclosed anywhere: the providers are named, their locations are not, so the jurisdiction applying to the stored data cannot be established from the published documents. The operator's own residence and jurisdiction are in Portugal, which is a separate matter from where the servers sit. The publisher reserves the right to store or process the data elsewhere in future, for strategic or economic reasons, without prior notice and subject only to an appropriate level of protection. Periodic backups are kept alongside the cloud servers and on local devices controlled by the publisher and his staff. Transmission security is not guaranteed, and the publisher advises against submitting sensitive data.
Things to keep in mind
Risks and trade-offs to weigh before adopting Promptmetheus.
- The subscription does not cover inference: your real monthly cost depends on your own provider API keys, and testing one prompt across dozens of models can easily outgrow the plan price.
- If you turn on key synchronization, third-party API keys leave your device for the database. The publisher explicitly declines responsibility for excessive requests caused by malfunctioning software or hacks, and requires you to set appropriate usage limits on those keys yourself.
- Every prompt you run is sent to the model provider you selected, which may store and process it under its own policy rather than this one — sensitive material leaves the tool's perimeter the moment you press run, and the publisher advises against submitting sensitive data.
- Cancel early and you lose the remaining time: no refund is issued for unconsumed days, and none at all outside an incorrect charge.
- Data you create during the 7-day trial may be lost if you do not go on to subscribe.
- No availability or error-free guarantee is given; the publisher reserves the right to remove content, accounts or the service itself, and in the event of a shutdown a pro-rata refund is only described as likely, never promised.
- One account per natural person and no sharing of subscriptions. More broadly, leaning on a tool that scores prompt variants for you is comfortable but can quietly erode your own sense of why a prompt works — keep reading the completions, not only the ratings.
Setup & Integrations
Technical difficulty
Low. Forge opens straight in a browser, with no installation and no account; Archery asks only for an email address and a password. Nothing is deployed and no software is installed — it is a pure web application, though it demands a 12-inch screen minimum. The one real hurdle is obtaining API keys from your LLM providers and pasting them in. No coding skill is needed to compose a prompt, but the block-and-variable logic assumes some technical literacy. Public documentation, a troubleshooting section and a Discord community back you up, and a 7-day trial lets you test before paying.
Deployment
Integrations
Supported languages
Behind Promptmetheus
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
What exactly is a Prompt IDE?
Which models and providers can I use?
Is inference included in the subscription?
Is there an API or an SDK?
Can I build and run AI agents with it?
What is the difference between Forge and Archery?
Does it integrate with Make, Zapier, IFTTT or n8n?
How is this different from the OpenAI or Anthropic playground?
What does it cost, and is there a free trial or a refund?
Who can sign up, how do I get help, and can I have my data deleted?
Should you pick Promptmetheus?
Promptmetheus is a niche tool, and deliberately so. It covers one stage of the LLM lifecycle — the individual prompt — and covers it with real depth: block-by-block composition, variations, automatic evaluators, completion ratings, cost estimates, and a version history that provider playgrounds simply do not keep. The publisher states the boundary himself: the tool designs, tests and optimizes prompts, it does not assemble or run agents, and the frameworks that do are treated as complements rather than rivals.
That narrowness is the honest part of the offer, and the thing to weigh before subscribing. If prompt quality is a genuine cost centre for you — an agent chain where errors compound, a team sharing a prompt library, a class working on prompting together — the depth earns its price. If you expected a platform that also ships the finished system, this is not it, and the missing API keeps it outside any automated toolchain for now.
Judging before paying is easy, which counts for a lot: Forge is free, needs no account and runs offline, and every paid plan opens with a 7-day trial. Budget the true cost, though — the subscription buys the workbench, not the inference, which is billed to your own provider keys on top.
Finally, weigh who is behind it. Promptmetheus is run by one individual in Portugal, privately owned and profitable by his own account, with a named DPO, subprocessors listed one by one and an explicit GDPR claim. Against that: no certification of its own, no published DPA, no disclosed hosting location and no stated retention period. For a personal prompt workbench that is a fair trade; for regulated data it is a conversation to have first.
- Choosing a selection results in a full page refresh.
- Opens in a new window.