zitejs, so
Zite delivers the context they need, when they need it.
The framework guide
Everycreate_sandbox response includes a framework guide: a concise spec of how Zite apps are
structured (monorepo layout, data access, zite.sql() identifier rules, access modes and sign-in, file
uploads, integrations and secrets, build workflow). Connecting the MCP and booting a sandbox is enough to
build correctly, with no memorizing conventions in advance. The same conventions are documented for
humans in the Framework reference.
The response also carries a docs index: the title, URL, and description of each framework, recipe,
concept, and deploy page on this site (plus Troubleshooting and Limits), plus docs.llmsTxt, the URL of the full index. The agent fetches
the Markdown pages it needs for the task in front of it, such as a recipe before building scheduled jobs
or Stripe payments.
llms.txt
This site publishes anllms.txt index and a full llms-full.txt so agents and IDEs can pull the docs into
context. Point your tool at:
Give your agent Zite context
When an agent works outside a Zite sandbox, building against the Database API or planning in your IDE before it connects, drop this into your project’sAGENTS.md (or a Cursor rule /
CLAUDE.md):
llms-full.txt, and link recipes directly (e.g.
/recipes/stripe-payments) for task-specific guidance.
Getting great results
- Be specific about data. Name the tables and fields; the agent designs the schema from your description.
- Ask it to verify at runtime. A clean type-check isn’t proof. See the build loop.
- Let it read
.zite/db.ts. It is the source of truth for table and field names. - Connect integrations first. Connect Slack or Stripe when the agent hands you a
setup_integrationlink, so the token is available while it builds.