Knowledge
Start with the question you have.
Three collections, three different jobs. This page exists to point you at the right one rather than make you read all of them.
Guides are procedural: open one beside your editor and follow it to a working setup. Articles argue a position and are meant to be read once, away from the keyboard — they are the case for doing this at all, not the instructions. Comparisons sit between the two: each takes one specific alternative, sets OrmAI beside it, and names the situations where that alternative is the better choice. If you are trying to decide, read a comparison. If you have decided, read a guide. If you are trying to convince somebody else, send them an article.
Three common starting points
- You want it running today. Take one of the 3 setup guides — OrmAI with Drizzle, OrmAI with Prisma, and Quickstart: OrmAI with FastAPI and SQLAlchemy — and pick by the ORM already in your codebase.
- You are comparing approaches. The comparisons cover hand-rolled tools, MCP database servers, raw SQL, and text-to-SQL. Each says where the alternative holds up, not only where it fails.
- You need to justify the work. The articles are the argument: what breaks at the model–database boundary, and why the usual answers do not hold under multi-tenancy.
Guides · 12
Procedures, in the order you need them.
The guides index groups these by level, because they build on each other: wire the policy to your existing ORM first, add redaction, tenant scoping, budgets, gated writes and the audit trail second, and review the whole thing against the production checklist last. Most carry runnable code for their stack.
- OrmAI with Drizzle
- OrmAI with Prisma
- Production checklist for agent + database systems
- OrmAI as an MCP server for Claude
- OrmAI with the Vercel AI SDK
- OrmAI with LangGraph
- Write operations with approval gates
- Query budgets and runaway agents
- Audit logs you'll actually trust
- Field-level redaction for PII
- Multi-tenant isolation for AI agents
- Quickstart: OrmAI with FastAPI and SQLAlchemy
Articles · 9
Field notes and positions.
These are essays rather than instructions. They cover the failure modes we keep finding in audits, the ideas worth borrowing from capability-secure operating systems, and one benchmark run. Read newest first; they do not depend on each other.
- Agent–database anti-patterns: a field guide
- Choosing between MCP, function calling, and policy runtimes
- Designing audit trails LLMs can't tamper with
- Capability-based security for AI agents
- From RAG to safe writes
- Capability tokens for AI: a primer
- The agent–data security gap nobody is talking about
- Why text-to-SQL fails in production
- Spider benchmark: 0 unsafe operations vs. 23 for text-to-SQL
Comparisons · 4
One alternative at a time.
One page each for hand-rolled tools, MCP database servers, raw SQL, and text-to-SQL. Rather than a feature matrix, each comparison takes a single one of those, describes the conditions under which it is genuinely the right answer, and then describes the point at which it stops being one.