diff --git a/README.md b/README.md
index d702a63..bdc4eb1 100644
--- a/README.md
+++ b/README.md
@@ -1 +1,17 @@
-# .usestring
\ No newline at end of file
+# .github
+
+Org-wide GitHub configuration for [@usestring](https://github.com/usestring).
+
+| Path | What it does |
+| --- | --- |
+| [`profile/README.md`](profile/README.md) | Renders as the org's public profile at [github.com/usestring](https://github.com/usestring). |
+| [`.github/workflows/gitleaks.yml`](.github/workflows/gitleaks.yml) | Secret scanning on pushes and PRs to `main`. |
+| [`.gitleaks.toml`](.gitleaks.toml) | Gitleaks ruleset. Loaded from the base branch on PRs, so a PR can't weaken its own scan. |
+
+GitHub reads the profile README only from a repo named exactly `.github`, at exactly
+`profile/README.md` — this repo was previously named `.usestring`, which is the *user*-profile
+convention (repo name == username) and gets no special treatment on an org.
+
+This is also where org-wide community health defaults belong (`CONTRIBUTING.md`, `SECURITY.md`,
+`ISSUE_TEMPLATE/`, `PULL_REQUEST_TEMPLATE.md`): any repo in the org that lacks its own copy
+inherits the one here. None are defined yet.
diff --git a/profile/README.md b/profile/README.md
new file mode 100644
index 0000000..346760c
--- /dev/null
+++ b/profile/README.md
@@ -0,0 +1,69 @@
+<h1 align="center">String</h1>
+
+<p align="center"><strong>Web access for everyone.</strong></p>
+
+<p align="center">
+  We build the web data API that doesn't get blocked — one request, any URL, clean content back.
+</p>
+
+<p align="center">
+  <a href="https://www.usestring.ai">Website</a> ·
+  <a href="https://portal.usestring.ai/docs">Docs</a> ·
+  <a href="https://portal.usestring.ai/sign-up">Get an API key</a> ·
+  <a href="https://www.usestring.ai/blog">Blog</a> ·
+  <a href="https://www.usestring.ai/careers">Careers</a>
+</p>
+
+---
+
+## The API
+
+Anti-bot handling, CAPTCHA solving, proxy rotation, and JS rendering are our problem, not yours.
+Send a URL, get the content back. A block costs you nothing — you're billed only when content is
+successfully delivered, and the first 5,000 requests are free.
+
+```bash
+curl https://request.usestring.ai/v1/fetch \
+  -H "Authorization: Bearer $STRING_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{ "url": "https://example.com", "format": "markdown" }'
+```
+
+Agents can skip the HTTP layer entirely — [`@usestring/mcp`](https://github.com/usestring/string-ai-mcp)
+is the official MCP server, local or hosted:
+
+```bash
+npx @usestring/mcp    # local; STRING_AI_API_KEY in the environment
+```
+
+...or point any MCP client at `https://mcp.usestring.ai/YOUR_API_KEY/v1/mcp`.
+
+## What we've open sourced
+
+|  |  |
+| --- | --- |
+| **[web-data-frontier-benchmark](https://github.com/usestring/web-data-frontier-benchmark)**<br>`TypeScript` | An open, reproducible benchmark for web access APIs: 90 real bot-protected URLs, 15 providers, one shared harness. We built it because [vendor success-rate claims don't survive a fixed test suite](https://www.usestring.ai/blog/web-scraping-benchmark-problem) — ours included, which is why the harness, the URL list, and the raw JSON are all public. |
+| **[powhttp-mcp](https://github.com/usestring/powhttp-mcp)**<br>`Go` · `AGPL-3.0` | X-ray vision into captured HTTP traffic for coding agents: TLS (JA3/JA4) and HTTP/2 fingerprinting, browser-vs-program diffing to locate detection vectors, API and GraphQL endpoint mapping, and response-schema inference. |
+| **[string-ai-mcp](https://github.com/usestring/string-ai-mcp)**<br>`JavaScript` · `MIT` | The official MCP server for the Web Access API — `web_access_fetch`, `web_access_search`, and quote-then-approve sitemap crawls, in any MCP client. |
+| **[utils](https://github.com/usestring/utils)**<br>`Shell` · `MIT` | The scripts and gists we reference in our writing. |
+| **[relay-self-hosted](https://github.com/usestring/relay-self-hosted)**<br>`JavaScript` | Fully self-hosted Obsidian live collaboration — CRDT sync, token service, and OAuth, with no relay.md control plane. |
+
+## The benchmark
+
+Official run — **2026-07-15** · 90 targets × 5 attempts × 15 providers = 6,750 requests.
+
+| Rank | Provider | Success rate | Passed |
+| ---: | --- | ---: | ---: |
+| 1 | **string** | **95.8%** | 431/450 |
+| 2 | scrapfly | 83.6% | 376/450 |
+| 3 | bright | 80.9% | 364/450 |
+| 4 | context_dev | 78.7% | 354/450 |
+| 5 | firecrawl | 70.9% | 319/450 |
+
+[Full table, all fifteen providers, and instructions to re-run it yourself →](https://github.com/usestring/web-data-frontier-benchmark)
+
+---
+
+<p align="center">
+  Built in New York City. &nbsp;·&nbsp; <a href="https://www.usestring.ai/careers">We're hiring.</a>
+</p>