Now your vibe-coding sessionscan actually ship to production

Short answer: a requirements.md is prose that only humans can reliably read; Entalpa is a structured, queryable source of truth that humans and coding agents share.

• Structure, not prose. Requirements in Entalpa are entities with stable ids (Req-012), acceptance criteria, rationale, and risk, plus explicit links: stakeholder → story → requirement → subsystem → interface contract. In a file, those relationships live in the author's head and silently rot.
• Agents read only what they need. Over MCP, an agent searches, filters requirements by story, subsystem, flag, or status, and batch-fetches exactly the ones it's implementing — instead of loading a 3,000-line file into context every session and hoping nothing important got truncated.
• Change tracking that understands requirements. Snapshots and diffs tell an agent precisely which requirements were added or modified since it last implemented anything — "implement only what's new" becomes one tool call. A git diff on a prose file can't tell a reworded rationale from a changed obligation.
• A closed loop. Agents write implementation status, code references, and test coverage back onto each requirement, and that progress is visible on the website. A file gives you no way to see, per requirement, whether it's implemented, verified, or blocked.
• One copy, many consumers. Product people edit on the website, agents work over MCP, and multiple repos can point at the same project — no forked copies drifting apart across repositories.

Short answer: it isn't a competitor — skills tell an agent how to work, Entalpa holds what to build. In fact, Entalpa ships as a skill plus an MCP server, because you need both.

A skill is a static instruction file: procedures, conventions, style. It has no live state — it can't hold your requirements, track which ones are implemented, link a requirement to its source story, or let a product manager edit anything. If you put your requirements inside a skill, you've just recreated the requirements-file problem with extra steps, and every edit needs a repo commit.

Entalpa is the stateful half: a database of stakeholders, stories, requirements, and subsystems with a web UI for humans and MCP tools for agents — search, diffing, traceability, and status feedback that no markdown instruction file can provide. The entalpa-implement skill then teaches the agent the discipline for using it: read before writing, preserve traceability, record implementation feedback. Use skills for process; use Entalpa for the requirements themselves.

Short answer: Superpowers improves how a coding agent executes a task (brainstorm → plan → TDD, inside one repo); Entalpa is where the requirements live — durable, structured, and shared with non-developers. They solve different problems and combine well.

Superpowers' artifacts are markdown plans and brainstorms committed to a repo: agent-facing, per-project, and frozen the moment they're written. There's no entity model, no stable requirement ids, no traceability from a stakeholder need down to the test that verifies it, and no interface a product owner or client can use without opening a pull request.

Entalpa sits upstream of that workflow. Stakeholders and stories are authored on the website, requirements are generated and refined there, and the coding agent — with or without Superpowers driving its execution style — pulls its work items over MCP, implements them, and reports status back per requirement. When the project changes, the agent diffs against a snapshot and picks up only what's new. A team can happily run Superpowers for engineering discipline while Entalpa stays the system of record for what the system must do and how far along it actually is.

Short answer: no training, and zero retention at the model layer. Your content is sent to an AI provider only to return your result — it's never saved by them and never becomes training data, for us or anyone else.

When Entalpa calls a model to generate or analyse requirements, your project content goes to a third-party AI inference provider — such as Anthropic, OpenAI, Google, Cerebras, or AWS Bedrock — purely to produce the result you asked for. In every case it is processed transiently: the provider does not store your prompts, and our agreements prohibit using your data to train models.

The structured spec you build does live in Entalpa — that's the point, so your agents and teammates share one source of truth. It stays private to your account, is encrypted in transit and at rest, and is never sold or used for training. You can export it any time (ReqIF, CSV), and on account deletion we remove your personal data within 30 days. In short: zero retention at the model layer, full control at the product layer.