Fabrik turns what your company already knows (products, prices, audiences, image rights) into blog posts, social posts, newsletters, videos and outbound emails. AI agents write from your facts, every draft is checked and approved by your team, and every piece is measured.
Free to set up. Your data stays in your workspace; nothing is published without an approval.
Rain at 4am, 900 m of climbing, and a shell that stays dry. The Ridgeline Trail Jacket is rated to 20,000 mm and weighs 310 g. €189, VAT included, in five colours.
Chat tools don’t know your catalogue, your current offers, which photos you may use in which country, or what your legal team banned last month. So someone rewrites every draft, checks every number, and hopes nothing slipped through.
Fabrik starts from your data instead of a blank prompt. It keeps a live model of your business, lets agents write only from it, proves every fact before a human sees the draft, and records who approved what.
Products, offers, audiences, assets. Through the app, the API or MCP.
Your business as linked objects: a product belongs to a collection, an offer targets a segment.
From facts and brand rules, per channel, language and market.
Prices, names, rights, language, links, brand. Failures are fixed once automatically.
Word-level diff, sources, cost, and who approved it, when.
Blog, social, email, outbound. The best pieces teach the next drafts.
Define object types and their properties, then drag a link between them: a Product belongs to a Collection, an Offer prices a Product in a Market, a Campaign targets a Segment. Every object keeps its history. Agents walk this graph to gather facts, and nothing else.
Every amount and offer name must match your catalogue. “From €149” when the price is €189 is caught.
Images and licensed material are checked by country, channel and date. A photo cleared for Europe never reaches a US ad.
Every product named exists in your ontology, with the right maker and specs. Misattributions are flagged.
The text is in the language you asked for, for every market you sell in.
Every link goes through a redirector with UTMs, so clicks and conversions are attributed to the piece.
One versioned document: voice, claims policy, banned terms. Change it and everything in review is re-scored.
And provenance on every version: source objects, prompt version, model, request id, cost. Enforced by the database, not by convention.
A DAM that knows where you may use things. Pick a market and a channel; anything not cleared is greyed out with the reason. AI tags every upload; web and social sizes are rendered automatically.
Three steps to a draft. Choose a format, choose what it’s about, choose the market and audience. The agent writes; checks run; a repair pass fixes what failed.
Iterate like it’s code. Every edit is a version; compare any two word by word. Approve, or reject with a reason the agent reads on the next try. Approval triggers publishing.
The three numbers that matter. Production time per piece, factual errors published (target: zero), and conversions attributed per piece. The best pieces become examples for the next generation.
Large catalogues, prices per market, image rights by season and country.
Titles, authors, editions and rights that differ by territory.
Plans and prices that must be exact everywhere, audiences by lifecycle stage.
One workspace per client brand, each with its own data, rules, team and reviewers.
Fabrik is an MCP server. Claude Code or Claude Desktop can search your ontology, read your brand rules, check rights and submit drafts for review, with the same checks as the factory. One command to connect.
claude mcp add fabrik --transport http \ https://byocontext.com/mcp \ --header "Authorization: Bearer fk_live_…" \ --header "X-Workspace-Id: …"