TECHNICAL WRITING
Ship technical docs in 40+ languages without breaking cross-references or code samples
Ellon AI translates technical manuals, API references, and knowledge base content while preserving Markdown and LaTeX formatting, cross-references, and code samples. output renders identically to the source.
Technical content breaks in flat-text translation
Technical writing isn't just prose. It's structured content with code samples, cross-reference links, command-line examples, and mathematical notation. A translator that treats the source as a blob of prose breaks the structure on every pass. Cross-references land in the wrong section. Anchor links resolve to nothing. Code samples inside fenced blocks get mangled. The tech writer spends more time fixing the translation than authoring the next version. Multiply that manual cleanup by the number of languages a product ships in and the number of releases per quarter, and translation becomes the bottleneck that makes localized documentation infeasible at most organizations. Ellon AI respects the source: code blocks stay verbatim, cross-references are preserved, anchor links hold, and mathematical notation in LaTeX or Markdown passes through unchanged.
- Markdown and LaTeX documentation formats supported natively
- Code samples, inline code, and command syntax never altered
- Cross-references, anchor links, and YAML frontmatter preserved
- Mathematical notation in .tex and inline math in Markdown intact
Translate a published manual page end-to-end
Upload a Markdown, LaTeX, Word, or PDF manual page. Ellon AI translates the prose while preserving CLI commands, cron expressions, YAML frontmatter, section numbering (§4.2 stays §4.2), cross-references ("see §5.3" follows the numbering), table structure, and navigation breadcrumbs. NOTE, WARNING, and TIP admonitions localise their labels while keeping the admonition styling, so the translated page renders identically to the source through your documentation toolchain.
Original · English
§4.2 Exporting telemetry data to cloud storage
TelemetryPro can export collected telemetry events to a managed cloud-storage bucket — Amazon S3, Google Cloud Storage, or Azure Blob Storage — on a schedule, on demand, or in response to alerting conditions. This section covers the scheduled-export path; for event-triggered exports, see §4.4.
Scheduled exports require a role with the telemetry.export permission granted at the workspace level. Workspace owners can grant this role from Settings › Permissions › Roles.
- 1Open the Export dashboard
From the main navigation, select Data › Exports. If you do not see the Exports item, verify that your role includes telemetry.export.
- 2Create a destination
Click New destination and provide a display name, the target bucket, and a service-account credential with write access. The credential is encrypted at rest.
- 3Define the schedule
Choose a cron expression or a preset cadence (every hour, every six hours, daily, weekly). Exports running on the preset cadence use the workspace time zone; cron expressions use UTC.
- 4Confirm and activate
Review the summary and click Activate. The first export runs at the next scheduled slot; you can trigger an immediate run with Export now.
Or, to configure the same export from the command line using the TelemetryPro CLI:
$ telemetry-pro exports create \
--name "prod-nightly-s3" \
--dest s3://my-bucket/telemetry/%Y/%m/%d/ \
--schedule "0 2 * * *" \
--format parquet \
--compression snappy
✓ Destination created (id: dst_8f2a1c)
✓ First run scheduled: 2026-03-13 02:00 UTCActivating a destination immediately enables writes to the bucket. Verify the target path and credential scope before activation — an incorrectly scoped credential may write telemetry data to an unintended project or account.
Combine scheduled exports with retention rules (see §4.3) to keep your operational dataset scoped while preserving a long-term archive in cloud storage.
| Format | Compression | Ideal for |
|---|---|---|
| JSON Lines (.jsonl) | gzip | Downstream analytics pipelines and custom loaders |
| Apache Parquet (.parquet) | snappy | Data-lakehouse workloads and columnar analytics |
| Avro (.avro) | deflate | Schema-evolving event streams |
| CSV (.csv) | gzip | Spreadsheet ingestion and legacy BI tooling |
Translated · Deutsch
§4.2 Telemetriedaten in den Cloud-Speicher exportieren
TelemetryPro kann erfasste Telemetrieereignisse in einen verwalteten Cloud-Speicher-Bucket exportieren — Amazon S3, Google Cloud Storage oder Azure Blob Storage — zeitgesteuert, auf Anforderung oder als Reaktion auf Alarmbedingungen. Dieser Abschnitt behandelt den zeitgesteuerten Export; für ereignisgesteuerte Exporte siehe §4.4.
Für zeitgesteuerte Exporte wird eine Rolle mit der Berechtigung telemetry.export auf Workspace-Ebene benötigt. Workspace-Eigentümer können diese Rolle unter Einstellungen › Berechtigungen › Rollen vergeben.
- 1Öffnen Sie das Export-Dashboard
Wählen Sie in der Hauptnavigation Daten › Exporte. Fehlt der Eintrag Exporte, prüfen Sie, ob Ihre Rolle die Berechtigung telemetry.export enthält.
- 2Erstellen Sie ein Ziel
Klicken Sie auf Neues Ziel und geben Sie einen Anzeigenamen, den Ziel-Bucket und einen Service-Account mit Schreibzugriff an. Die Zugangsdaten werden verschlüsselt gespeichert.
- 3Definieren Sie den Zeitplan
Wählen Sie einen Cron-Ausdruck oder ein voreingestelltes Intervall (stündlich, alle sechs Stunden, täglich, wöchentlich). Vordefinierte Intervalle verwenden die Workspace-Zeitzone; Cron-Ausdrücke laufen in UTC.
- 4Bestätigen und aktivieren
Prüfen Sie die Zusammenfassung und klicken Sie auf Aktivieren. Der erste Export läuft zum nächsten geplanten Zeitpunkt; mit Jetzt exportieren lösen Sie einen sofortigen Lauf aus.
Alternativ konfigurieren Sie denselben Export über die TelemetryPro-Befehlszeile:
$ telemetry-pro exports create \
--name "prod-nightly-s3" \
--dest s3://my-bucket/telemetry/%Y/%m/%d/ \
--schedule "0 2 * * *" \
--format parquet \
--compression snappy
✓ Destination created (id: dst_8f2a1c)
✓ First run scheduled: 2026-03-13 02:00 UTCDie Aktivierung eines Ziels erlaubt sofortige Schreibzugriffe auf den Bucket. Prüfen Sie vor der Aktivierung den Zielpfad und den Umfang der Zugangsdaten — falsch konfigurierte Zugangsdaten können Telemetriedaten in ein falsches Projekt oder Konto schreiben.
Kombinieren Sie zeitgesteuerte Exporte mit Aufbewahrungsregeln (siehe §4.3), um Ihren operativen Datensatz schlank zu halten und dennoch ein Langzeitarchiv im Cloud-Speicher zu führen.
| Format | Komprimierung | Ideal für |
|---|---|---|
| JSON Lines (.jsonl) | gzip | Downstream-Analyse-Pipelines und eigene Loader |
| Apache Parquet (.parquet) | snappy | Data-Lakehouse-Workloads und spaltenbasierte Analysen |
| Avro (.avro) | deflate | Ereignisströme mit sich entwickelndem Schema |
| CSV (.csv) | gzip | Tabellenkalkulations-Ingest und ältere BI-Tools |
Hand the localisation team a full release delta
Upload the old and new version of a DOCX or PDF export of your manual or reference. Ellon AI produces a Word tracked-changes document covering every section that moved. renamed navigation paths, updated procedure steps, default-value changes, deprecated CLI flags, new admonitions, new format rows. plus a migration checklist the localisation team can execute against every language variant. Designed for release engineering teams keeping a multi-language documentation set in lockstep with the English master.
This revision lists every chapter the localisation team must propagate between Platform User Guide versions 3.8.2 and 3.9.0. Items flagged breaking require a parallel change in every language variant; others are additive. Code snippets, CLI commands, and cron expressions pass through unchanged.
The scheduled-export procedure now starts from Settings › Exports Data › Exports, a navigation move that applies to every language variant. The procedure wording has been tightened:
Step 3 — Choose a cron expression or a preset (hourly, 6h, daily, weekly). Preset schedules use workspace time; cron uses UTC. Step 3 — Choose a cron expression or a preset cadence (every hour, every six hours, daily, weekly). Exports running on the preset cadence use the workspace time zone; cron expressions use UTC. The preset cadences now honour daylight-saving transitions automatically.
Retention defaults have changed and must be reflected in every translated table:
The default retention window for exported telemetry is now thirty (30) days ninety (90) days, with an optional seven-day grace period, aligned with the retention policy change described in §4.3.1. A new subsection §4.3.2 covers legal-hold exceptions, which must be called out in every translated variant: legal holds suspend automatic deletion regardless of the configured retention window.
A new section §4.4 documents event-triggered exports — a workflow where a webhook or internal alert fires an immediate export of a scoped time window rather than waiting for the scheduled slot. The section includes a new procedure (five steps), two new admonitions (NOTE and WARNING) and a new CLI example. Translators should mirror the admonition labels in the target language per the style guide (NOTE → HINWEIS, WARNING → WARNUNG, TIP → TIPP).
Two CLI flags have been deprecated in 3.9 and should be marked as such across every translated CLI reference:
--retention-days DEPRECATED — use --retention-window "90d" --compression-tier DEPRECATED — use --compression snappy | gzip | deflate
The supported-formats table gains one row and loses another:
Added — Apache Iceberg (.iceberg) with zstd compression, ideal for lakehouse workloads with schema evolution. The table row order should be JSONL, Parquet, Iceberg, Avro, CSV in every language version. Removed — the legacy fixed-width (.fw) format is no longer supported and the row should be dropped from every table.
A new WARNING admonition is required at the top of §7.4: “Workspace administrators granting telemetry.export to a new role should also enable the audit-log feed on that role — failure to do so masks who triggered an export in the activity timeline.” Translators should match the existing WARNING style used elsewhere in §7.
Before publishing any translated 3.9 variant, the localisation team must: (i) update every navigation path reference from Settings › Exports to Data › Exports; (ii) update default retention windows from 30 days to 90 days in every table and procedure; (iii) add the new §4.4 and §4.3.2 sections to the TOC; (iv) add the new WARNING in §7.4; (v) run the docs-diff verification pipeline to confirm no missed strings.
All other 3.8 manual content carries over unchanged — Installation, the API reference, the SDK reference, the cookbook, and the troubleshooting guide. Full propagation coverage is expected in every locale within 5 business days of the English 3.9 release.
How technical writing teams use Ellon AI
Technical writing at scale is a documentation factory. A large SaaS platform might maintain a set running to thousands of topics across user guides, API references, release notes, tutorials, and knowledge base articles. and that entire corpus usually needs to be available in at least a handful of languages for international customer segments. Traditional translation vendor workflows handle volume poorly: every content update in the English master triggers translation work in every language, and the coordination overhead eats the writing team's productive capacity.
Product documentation at scale
Translate user guides, administrator documentation, and feature references across the languages your customer base operates in. Ellon AI handles the volume in batches that process in hours, not weeks. Markdown, Word, and PDF source formats are supported; the translated content builds through the same documentation toolchain as the source.
API references and developer documentation
Long-form API references, SDK documentation, and protocol specifications are code-heavy and cross-reference-heavy content types. Ellon AI preserves code samples verbatim across translation, maintains anchor links and cross-references, and handles the structural formatting that API documentation frameworks (Docusaurus, VitePress, GitBook) rely on.
Release notes and changelogs
High-frequency documentation like release notes produces continuous translation demand. every release means new content across every supported language. Ellon AI handles this cadence natively; release notes translate in minutes, so the release announcement can go out simultaneously in every language rather than staggered by multi-day translation queues.
Knowledge base and support articles
Customer support knowledge bases are often the most-viewed documentation a product has, and translation into customer languages materially affects support volume and customer satisfaction. Ellon AI handles KB content at scale. within each article, terminology stays consistent because the translator builds a per-document glossary automatically.
Tutorial and onboarding content
Product tutorials, getting-started guides, and onboarding content translate with code samples intact and interactive elements preserved. For product-led growth organizations that rely on self-serve onboarding, having the onboarding content in the customer's language is a direct conversion lever.
Release verification
When a release ships translated documentation alongside the English master, the compare tool on the old and new English source produces a Word tracked-changes document with AI semantic analysis. new features, deprecated parameters, and workflow changes flagged separately from formatting and terminology shifts. That change list is what the localization team propagates across translated versions.
Terminology management
Product terminology, UI string vocabulary, and branded product names need to translate consistently within each document. The translator automatically extracts a per-document glossary from the source and enforces it across the output. For cross-document consistency, compare translations against a reference document or verify via the compare tool between translated outputs.
Site and platform internationalization
For documentation sites and developer platforms that offer i18n-aware hosting, Ellon AI handles the content translation layer while the hosting platform handles locale routing, fallbacks, and language switching. Translated content lands in the expected locale-specific directory structure or CMS locale, ready for publication.
Review and quality workflows
Translated documentation typically goes through a human review step before publication. either in-house reviewers in the target language or external review vendors. Ellon AI's output is designed as a high-quality baseline that accelerates review rather than requiring it from scratch.
Format coverage
Markdown (with GitHub and CommonMark extensions), LaTeX, Word (.docx), and PDF are supported. YAML frontmatter is preserved. Structural elements. headings, lists, tables, code blocks, cross-references. pass through unchanged.
Language coverage
Ellon AI supports 200+ language pairs, including the common technical writing inventory: English to/from German, French, Spanish (regional variants), Portuguese (Brazilian and European), Japanese, Korean, Chinese (Simplified and Traditional), and Russian. Technical terminology consistency holds across these pairs within each document.
The net effect: technical writing teams that used to choose between documentation coverage and documentation completeness can now have both. The same documentation set ships in every customer language on the same cadence as the English updates.
Frequently asked
Frequently asked questions
Still have questions?
Reach out and we'll answer within one business day.
Ship your next release in every customer language
Start with 20 free pages. No credit card. Upload an API reference and see the cross-references and code samples survive the translation.
Start your free trial