AI can already write. The hard part is shipping it safely.

If you build with AI on top of your own content, you already know the model is rarely the bottleneck. The hard part is putting it to work across a real site without breaking production: scattered context, improvised scripts, review happening in the wrong place, and a publish step that’s still fragile manual hops. Your archive lives behind an editor and a render pipeline, and the obvious move — pointing AI at one post at a time, or scraping your live HTML — gives you the worst possible substrate and no safety net.

Specter is the workflow and control layer around that work. Connect your site, point a recipe at a job with full-site context, and review every change as a diff before anything goes live. Subscribe now for 500 free credits.

Full-site context, then a diff you can trust

The reason a CMS editor feels small is that it only ever shows the model one post. Specter runs AI as recipes across the whole connected archive — an internal-linking pass that reads every post and proposes links between related pieces, a rewrite that matches the tone of the surrounding site, an answer-readiness audit that flags posts missing clear answers or structure. Then it shows you exactly which records would change before it ships. Browsing and reviewing cost nothing; only AI runs spend credits. The intelligence does the work; the review step is what makes it safe to run against a live site. Using AI to edit posts walks through how that loop feels.

A caveat worth being plain about

The editable content is a projection of the underlying post, not the post itself — some platforms (Ghost especially) store content as Lexical, and card-heavy posts with embeds, galleries, or custom HTML don’t always round-trip losslessly. For reading and analysis the prose comes through clean; it matters when you write back. That’s what the dry-run is for: see the blast radius of an AI pass before it touches production, and slow down on card-heavy posts. How Specter handles Ghost cards is the full accounting of what survives.

Building RAG, MCP, or embeddings? Use the local corpus

If what you actually want is your content as clean files to chunk, embed, and diff — a local corpus for retrieval, an MCP server your agents call, a Custom GPT, an llms.txt you keep current — that’s the desktop and open-source edition. It two-way syncs your whole archive to a folder of plain .md on disk: no scraping, no HTML cleanup, no rate limits against your own production site, and frontmatter you can carry straight into metadata. Specter bundles no model and runs no embeddings — it gives you trustworthy local markdown and the round-trip back, and gets out of the way. The intelligence and orchestration are yours; the corpus stays current.

The promise is narrow on purpose: put AI to work across your whole site with a diff in front of every change — and, when you need it, a clean local corpus to build on.