Gå til hovedinnhold

Kapi

Kapi holds your project's context — the terms, the voice and the rules it actually goes by — where your tools and your AI can read it. It parses content in any format, resolves what applies to a given piece of it, and writes the original back byte-for-byte. It is built on the open-source neokapi content engine.

Why a project needs one​

Communication is contextual. You do not write a legal notice the way you write a help article, or a release note the way you write a landing page — same product, often the same fact, but the register, the vocabulary and how much is assumed of the reader all move with the audience and the surface.

Most projects hold that informally: a style guide nobody opens, the judgement of two or three people, a tacit sense of that is not how we say it here. That was workable while only people wrote. It is not workable now that agents write a growing share of the output and have no way of knowing where they are.

So kapi keeps it in a form a machine can read, and answers one question for three different callers — the AI agent asking what applies here, the check asking does this rule fire, and the engine asking may this approved wording be reused.

What it does with it​

  • Keep your source content right — edit it faithfully and verify it against the context that holds there, with checks that act like tests for AI output (a draft scores 0–100 and a gate passes or fails).
  • Say it the same way everywhere — one verb, kapi up, brings every file and every language up to date: draft with the AI you already have, reuse what you have already approved, enforce your terminology, and park what needs a person for review — all without ever mangling the file's structure.

A language is one more axis of the same context: staying on-profile in Norwegian is the same question as staying on-profile in a help article. Nothing here needs a second language to be worth having.

Three ways to run it​

Kapi ships as one binary, an agent skill and an MCP server — free forever. Every surface works the same engine, tools, flows, and .kapi projects, so a workflow built in one runs in the others:

  • Kapi CLI — a single binary that operates directly on files. Best for automation, scripting, and CI.
  • Your AI agent — the agent skill and the MCP server are how the context reaches whichever AI you already use: Claude, or any skill/MCP-capable tool, driving kapi by plain language inside its guardrails. See Use with Claude.
  • Kapi Desktop — a visual app: create projects, watch coverage against the ship gates, review and approve translations, build flows, manage plugins and credentials — no YAML, no flags.

Not sure which fits? See Kapi CLI or Kapi Desktop?

How you'll work​

You drive Kapi either ad-hoc (one-off, on files you name) or as a project (a kapi.yaml recipe plus a store that remembers your work) — the same commands, a different binding. See Projects.

From there it's task by task: the Kapi CLI recipes walk preparing your source and going multilingual, and Kapi Desktop does the same work visually.

Next​