How it works

From repo to first edit.

Five steps. About two minutes of setup. After that, every routine edit is one chat message away — and your developer only sees the real work.

01Connect

Connect your repo

Install the CMS Chat GitHub App on the specific repo you want to edit. We never see your other repos. The install grants read access to the codebase and write access to one branch you control.

On install, CMS Chat detects the framework — Next.js, Astro, Hugo, Nuxt, plain HTML, anything in Git. No plugins. No migration. Your repo stays exactly the way it is.

Connect repo

Repository

GHacme-co/websiteInstalled
✓

Next.js 16 detected

App router · MDX content

02Scan

CMS Chat maps your editable zones

An automated scan walks the file tree, identifies content files (TSX, MDX, Markdown, HTML, JSON), and tags each editable region. Headlines, body copy, prices, image references, links — all mapped automatically.

Most projects finish in under 10 seconds. The output is a list you can review: every component, every content file, every editable zone. Nothing is enabled until you say so.

Editable zones found

components/Hero.tsx4 zones
components/Pricing.tsx6 zones
content/about.mdx12 zones
content/blog/*.mdx24 zones
components/Footer.tsx2 zones
48 editable zones mapped in 6.2s
03Guardrails

Define safe zones and SEO rules

Pick what’s safe to edit. Text, images, prices, links — usually safe. Layout, code, scripts — usually locked. SEO meta and H1s usually live in “warn” mode so editors see ranking impact before pushing.

Connect Google Search Console (optional, Pro plan) and we’ll pull live position + traffic data for every page. Now warns are concrete: “this H1 ranks #3 for X.” Configure once and forget it.

Guardrail config

Text contentEditable
Images & alt textEditable
Pricing valuesEditable
SEO meta tagsWarn
Page layoutEscalate
Code & scriptsLocked
04Chat

Edit by chatting in plain language

Open the portal — yourcompany.cmschat.ai or your custom subdomain. Type the change you want. The AI finds the file, writes the edit, runs the guardrail check, and commits to your branch.

Routine edits go live in 4–8 seconds. Warned edits get a confirmation step. Locked content is refused with an explanation. Escalations open a PR for your dev.

edit.acmeco.com
Update the price on the Pro plan to $59
Updated Pricing.tsx line 24. Committed to main.
Live · 4s
05Escalate

Big changes route to your developer

Anything that touches structure, layout, code, or genuinely complex logic doesn’t get committed automatically. CMS Chat opens a real GitHub pull request with the proposed change and notifies your team.

Your developer sees a normal PR — diff, description, branch — reviewable in their normal workflow. They merge when ready, or comment back asking the editor to refine the request.

Routed to dev — PR #142

Add testimonials section to /home

Requested by Sarah · 2h ago

+ 38 - 4
components/Testimonials.tsx
app/page.tsx
Architecture

What happens behind the scenes

Every edit goes through the same pipeline. Same checks for the owner, the agency, the client portal user.

Edit pipeline

01

Editor

Owner / team / client

02

CMS Chat

Routes by zone

03

Guardrails

Zone + SEO check

04

GitHub

Real Git commit

05

Website

Live in seconds

Safe

Commit straight to main

Warn

Show SEO impact, ask

Escalate

Open PR for the dev

Want a deeper view? The home page has the live escalation flow diagram with animated edit paths. See it in motion →

Technical FAQ

For the people doing the integration

Repo access, frameworks, models, security. The questions developers actually ask. Email us for anything missing.

What does CMS Chat actually access in my repo?
We install as a GitHub App on the specific repos you choose. We can read those repo’s contents, write to a configured branch, and open pull requests on them. We never touch other repos in your account or organization.
Where do my changes live?
Every edit is a real Git commit on your branch — usually main or a configured branch. We don’t store your content in a separate database. Cancel CMS Chat and your site keeps running unchanged.
Which frameworks do you support?
Anything in a Git repo. We auto-detect Next.js, Astro, Hugo, Nuxt, Eleventy, Gatsby, Docusaurus, Mintlify, and plain HTML/Markdown. Custom stacks usually just work — the AI reads the file structure and adapts.
Which AI models do you use?
GPT-4o-mini by default for cost-efficient routine edits. Larger Anthropic and OpenAI models are available on Pro and Studio. You can also bring your own OpenAI or Anthropic key on those tiers.
How do guardrails actually work?
Each editable file or component gets a zone tag — editable, warn, escalate, or locked. The AI checks the tag before any write. Locked content can’t be touched. Warn content shows a confirmation. Escalate content opens a PR.
Is my repo content sent to model providers?
Yes — only the files relevant to a given chat turn are sent to the configured model provider for inference. We don’t train on your content. Bring-your-own-key bypasses our infrastructure entirely.
Do you support self-hosting?
Not in the public product yet. Studio customers can request a self-hosted deployment with their own infrastructure and their own model keys.
How does SEO Guardrails know what ranks?
We connect to your Google Search Console (read-only) and pull keyword, position, and traffic data per page. Before any change to an H1, title tag, or meta description, the AI surfaces the at-risk metrics.

Stop emailing your developer for typos.

Connect your site in 60 seconds. Make your first edit by chatting. Your developer handles the real work.

Start for freeSee how it works →