The Arc documentation has been rebuilt from the ground up. We moved off Docusaurus and onto Fumadocs. It is live now at docs.basekick.net.
This is not a redesign. It is a different stack, a different information architecture, and a different set of assumptions about who reads documentation in 2026.
Why we moved
Docusaurus served us well through the first year of Arc. It is a solid, mature project and it got us from zero to a real documentation site fast.
The problems that pushed us off it were specific:
No search. This was the most common complaint we got, by a wide margin, and the embarrassing truth is that the old site had no search box at all. Docusaurus does not ship search by default: you either apply for Algolia DocSearch and depend on their crawler and index, or you bolt on a community plugin. We never wired up either, so finding anything meant knowing the sidebar by heart or asking in Discord. For a database with a configuration surface as wide as Arc's, that is not a documentation site, it is a filing cabinet.
Structure the tool owned, not us. Docusaurus generated category landing pages from sidebar labels and slugified the label to make the URL, so some URLs did not match the directories the content lived in (/arc/category/cli-arcctl/ came from docs/cli/, for example). Every release added pages, and the generated structure drifted further from the shape of the content. We also still carried full doc trees for two retired products, because removing them cleanly was never trivial.
LLM readability. Increasingly, people do not read our docs. Their agent does. Someone asks Claude or Cursor how to configure MQTT ingestion in Arc, and the quality of that answer depends entirely on whether our docs are structured in a way a model can consume cleanly. That was not something Docusaurus was built for, and it is not something we could bolt on well.
What changed
Search that works
Search is built in and instant. Hit ⌘K from any page.
Under the hood it is Orama. The full-text index is built at build time and ships with the site as a static asset, so search runs entirely in your browser. There is no hosted service, no API keys, no crawler, and no indexing lag: the moment a deploy goes out, every new page is searchable.
Three doc sets, cleanly separated
Arc OSS, Arc Enterprise, and Arc Launchpad are now three distinct doc sets, each with its own navigation and its own authored landing page. The doc trees for two retired products are gone (their URLs return an honest 410 instead of a broken page).
This matters practically. If you are running the open source binary on a single node, you should never wander into cluster RBAC configuration you do not have access to. If you are operating an Enterprise cluster, you should not have to guess which parts of a page apply to you.
- Arc OSS: install, ingest over Line Protocol or MessagePack, query with SQL, connect Grafana, Telegraf, MQTT, or OpenTelemetry
- Arc Enterprise: multi-node clustering, RBAC, audit logging, tiered storage, query governance
- Arc Launchpad: the self-hosted console: SQL, logs, tokens, retention, alerts, continuous queries
Built for agents, not just people
Fumadocs generates machine-readable output alongside every human-readable page, and all of it is live today:
/llms.txt: an index of every page with its title and URL, the entry point for an agent exploring the docs/llms-full.txt: the entire documentation corpus as one Markdown file, for tools that want everything in context at once- Every page is also available as raw Markdown at
/llms.mdx/<path>/content.md, and every page has a Copy Markdown button so you can paste clean source straight into a conversation with your agent
The practical effect: when someone points an agent at the Arc docs, the agent gets clean structured content instead of scraped HTML with navigation chrome mixed in.
What did not change
Every URL that existed before the migration still resolves. We treat URLs as a contract: the full list of pre-migration URLs lives in the repo, the build fails if any of them stops resolving, and the deploy pipeline runs the same check against the live site after every push. The old auto-generated category pages now redirect to real landing pages at their natural URLs. If you have deep links to the Arc docs in a runbook, a wiki, or a bookmark, they still work.
The content is the same content. This migration was about the platform, not a rewrite. Pages that were accurate before are accurate now, and the gaps that existed before still exist. We are working through those separately.
What is next
The ASCII architecture diagrams are next. While converting the content we found 69 fenced blocks full of box-drawing characters. Most of them are legitimately monospace (terminal output, directory trees, code samples) and will stay that way. But 23 of them are pictures drawn in characters: the WAL write path, tiered storage, the compaction lifecycle, the RBAC hierarchy, the AWS reference VPC. We inventoried every one during the migration and are replacing them with real diagrams, drawn as SVG so they stay sharp and follow your light or dark theme, plus actual screenshots for the pages that today render Grafana and VS Code interfaces in ASCII art.
If you find something wrong, missing, or confusing, open an issue on GitHub or say something in Discord. Documentation bugs are bugs.
Arc is an open source, SQL-native time-series database. Single Go binary, Parquet storage on infrastructure you own. Get started.