Telemetry chose Blume for its first docs.

Telemetry helps developers find failures, slow calls, and unexpected costs in AI applications and agents. An AI request can pass through several model calls, tools, and retrievals before it produces an answer. Telemetry brings those steps into one trace, with latency, errors, token usage, and cost estimates, so developers can see where a request slowed down or failed.

Telemetry is open source and built on OpenTelemetry. It works with its own TypeScript and Python SDKs or an existing OpenTelemetry setup, and its docs cover all of it: a quickstart, both SDKs, the ingest API, and a guide for each integration, from the Vercel AI SDK and LangChain to Claude Code and Cursor.

Starting from scratch

Telemetry had no docs to move. Getting them started meant making a few choices early: where they would run, how much work hosting them would take, and how well coding agents could work with them. Blume fit all three, and became Telemetry's first docs platform.

Ephraim Atta-Duncan, Telemetry's co-founder, committed the first version on August 7, 2026: 29 pages covering the SDKs, the platform, the ingest API, and the first 11 integrations.

Hosted where Telemetry chose

The first thing that stood out was control. Blume is open source and free to use, and it builds docs into static files that any host can serve. Being free mattered, but being easy to self-host is what made that freedom practical.

Telemetry serves its docs from Cloudflare as static assets, with no server to run. They started at docs.telemetry.dev, and a month in, the team moved them next to the product at telemetry.dev/docs. That took two settings: a base of /docs in blume.config.ts, which moves the whole site under the path, from every page to its sitemap, llms.txt, and Open Graph images, and a route for the path in Cloudflare. Old links to the subdomain redirect to the new address.

Built to work with agents

Telemetry helps developers understand what happens inside AI applications and agents, so a docs framework that works well with agents made sense for the product. Blume's CLI mattered as much as where the docs could run.

Blume's CLI reports problems where an agent can act on them: blume build fails on a page that doesn't parse instead of shipping without it, and blume doctor and blume validate print JSON for an agent to read. Blume's agent skills teach Claude Code, Codex, and Cursor how a Blume site fits together.

An integration hub

The docs grew with the product. Telemetry has added 9 integration guides since launch, for 20 across frameworks, model providers, and coding agents, and the docs now start from them: send a first trace, then pick the integration that matches your stack.

Two component overrides carry each integration's logo through the docs: one puts it beside the page title, and an integration card built on Blume's own Card shows it on the overview grid. The rest of the look comes from Blume's theme options, a black-and-white accent and a medium corner radius, plus 36 lines of CSS for the logos.

Agents get the same hub without the markup. Every page has a Markdown version, and the site publishes llms.txt and llms-full.txt. A serializer in blume.config.ts turns each integration card into a plain Markdown link with its one-line description, so an agent reading the overview gets a clean list of all 20 guides.

The results

Telemetry's docs have been on Blume since the first page. They run as static files on Cloudflare at telemetry.dev/docs, next to the product, and every page has a Markdown copy for agents.

The docs are live at telemetry.dev/docs, and their source is public on GitHub. To start docs of your own, follow the quickstart.