Getting started
Lyntcast sits between AI agents (and people) and your CMS. Agents propose changes; Lyntcast diffs them, checks them against your rules, scores the risk, and then applies, holds or blocks them.
The five ideas
- Workspace: your team, its sites, rules, agents and audit log. Members have a role: owner, admin, reviewer or viewer.
- Site: one connection to a CMS (a WordPress site, a Sanity dataset, a Shopify store…). Lyntcast stores the credentials encrypted; agents never see them.
- Agent: an AI client or automation allowed to read and propose changes. Each agent has a trust level, optional daily limits and can be paused or revoked.
- Rule: a check every change goes through, such as a price limit, a protected field or a banned phrase. Each rule flags, requires approval or blocks.
- Change: one proposed create, update or delete. It ends up applied, pending review, scheduled, blocked, rejected, undone or in conflict.
Set up in four steps
- Sign up. Create an account. You get a workspace with a demo store, the Essentials rule pack and a first agent key, so you can try the full flow before connecting anything real.
- Connect a site. Go to Sites and add your CMS. WordPress connects in one click; other CMSs need one API token. See Connect your CMS for exact steps. Lyntcast checks that it can write before saving the connection.
- Review your rules. Under Rules, install packs (Essentials, E-commerce, Financial services, Health and wellness) and adjust limits. See the rules reference.
- Connect your agent. Under AI agents, copy the MCP URL into Claude, ChatGPT or Cursor, or create a key for your own agent. See Connect your AI agent.
What happens to a change
- Diff. The proposal is compared with the document’s current state, field by field, including rich text.
- Rules. Every enabled rule runs. Each finding carries an action:
flag(medium risk),require_approval(high) orblock(critical). - Risk and decision. The agent’s trust level decides what goes live without a person. Blocked changes are never applied; the agent gets the reasons.
- Apply. Before writing, Lyntcast takes a snapshot of the document so the change can be undone.
- Record. Every step is written to a hash-chained audit log.
Trust levels
| Level | Goes live without review |
|---|---|
strict | Nothing. Every change waits for a reviewer. |
standard | Low-risk changes. |
trusted | Low- and medium-risk changes. |
adaptive | Starts strict; becomes standard after 20 clean approvals and trusted after 100, and steps back down if more than 15% of recent reviews are rejections. |
Creates and deletes are never low risk, and a newer proposal touching the same fields replaces an older pending one.
Brakes
- Emergency stop pauses every agent in the workspace at once. Any agent can also be paused on its own.
- Daily limits per agent cap changes and deletes in a rolling 24 hours (defaults: 500 changes, 20 deletes).
- Automatic pause when an agent suddenly makes a burst of changes or many of its recent changes are rejected, a sign it may be following instructions it read somewhere it shouldn’t.