Gå til hovedinnhold

Guides

Each guide takes a single task — check content, translate it, review what a machine couldn't decide — and shows the same job through three doors, in order: Kapi Desktop, your AI agent, and the CLI. The interactive embeds run the real kapi binary in your browser: no install, no server, no API key.

For what kapi is and how the surfaces relate, see the Kapi overview. For every command and flag, see the command reference.

Get oriented​

kapi runs entirely in your browser here — no install, no server. Three commands orient you: kapi --help lists the top-level commands, kapi formats enumerates the format readers, and kapi tools lists the processing tools you can run on extracted content. Step through the tour from the rail, or press Run all.

Loading the walkthrough…

The explorer below shows the format engine round-trip: run a file through pseudo-translate and compare the source with the output — only the leaf text changes, while the surrounding skeleton (tags, keys, attributes) returns byte-for-byte. This is the same reader and writer the recipes drive.

Loading the interactive lab…

Ad-hoc or project?​

A .kapi project is the recommended working model for any repository or content set you work on repeatedly: languages, content patterns, and flows in a committed recipe, git-style discovery from any subdirectory, and a store that accumulates content memory, terminology, and approvals across runs. Create your first project walks the setup; Projects is the concept. kapi also runs ad-hoc — a command on the files you name, nothing saved — the quick path for one-offs, scripting, and a fine on-ramp. Guides note the project form where it differs.

Prepare your source​

Feed content to your AI

Parse any format into anchored content blocks an assistant or RAG pipeline can read.

Edit content

Edit the text inside any file — you supply the edits, kapi writes them back through the faithful round-trip.

Create content

Author from scratch with kapi as the checker — parse it back, gate on brand and terms, revise.

Check content like tests

Structured findings, a pass/fail gate, an exit code CI understands — and check --ship as the release bar.

Keep source on brand

Flag forbidden, competitor, and off-voice terms in your source and rewrite the ones that fail.

Set up your brand guide in 10 minutes

From starter pack to a bound, enforced brand voice profile.

Content governance for AI

Brand profile + terms + checks as guardrails, so agent output arrives already compliant.

Bulk rewrite with an agent

Sweep one editorial change across a whole content set, structure guaranteed, completion proven.

What gets translated

Protect sensitive content, reuse past translations, and keep terms consistent — one pass before anything ships.

Go multilingual​

Translate with your AI

kapi up catches a project up to its ship gates; kapi translate does one file — formats round-trip intact.

Review & approve

The human half of the loop: work the review queue, approve or reject, raise the coverage a gate measures.

Choose a provider

Select an LLM or MT backend, wire its credentials safely, and make it the default — including on-device Ollama.

Estimate a run's cost

Count blocks and source words in any format to size a job before you commission it.

Reuse what you've translated

Leverage existing content memory, fill the gaps, and pre-flag terminology — deterministic, no API key.

Keep terms consistent

Import a term list and flag terminology drift in translated files before they ship.

Preview before you ship

Pseudo-translate your strings to surface clipping and missing-string bugs before real text arrives.

Translate audio & video

Pull text out of audio, video, and images, translate it, and write subtitles or captions back.

Ship gates & CI

Enforce the project's release bar on every PR with kapi check --ship.

Tier gates per market

Contractual locales human-reviewed, the long tail machine-shipped — one recipe encodes the review budget.

Pause & resume a big run

A single-file .kpz workspace: extract in, transform in place, merge out — config travels with the file.

Where translations & terms live

The three storage models — loose file, named store, or project store — and the flags that select each.

Hand off to translators

Extract to XLIFF or PO, hand it to a translator, merge the targets back — with full memory provenance.

Next steps​