DynoTable
Blog

DynoTable Engineering Blog

How DynoTable is built — the engineering stories behind the AI agent, the MCP server, and the rest of the product.

These are working notes from building DynoTable, a desktop DynamoDB client. Each post walks through one engineering problem we shipped through for real — staging every write behind a reviewable diff, running an OAuth server inside an Electron app, making a local SQL engine respect DynamoDB’s access-pattern rules.

We write them from the commits and the design notes — expect concrete constraints, the decisions we reversed, and code where it matters.

If you want DynamoDB fundamentals rather than build stories, the Learn section is the structured course — indexes, single-table design, capacity math and the rest.

· 10 min read

How many writes per second before a new DynamoDB table throttles?
We offered a new on-demand DynamoDB table 8,000 writes/s. It served 4,000 and throttled the rest, and more capacity took 8.6 minutes to arrive. Data, method and the warm-throughput fix.

· 7 min read

Why we hand-wrote a PartiQL parser for DynamoDB
Why we hand-wrote an error-tolerant PartiQL parser for DynamoDB's dialect, with source offsets for quick fixes, depth caps and 103 fixtures that cite AWS docs.

· 10 min read

Your agent doesn't need every tool in its context
How we decide which tools an LLM agent sees up front and which wait behind a search tool, and why cheap models made us widen the visible set.

· 14 min read

GPT-6 Luna scores within error of Nova 2 Lite at 1/8 the cost
On 189 DynamoDB agent eval cases, GPT-6 Luna showed no detectable gap to Amazon Nova 2 Lite at about 1/8 the cost per trial. Plus GPT-5.6 Luna, a 12-model ladder and caching.

· 10 min read

How we built a local SQL engine for DynamoDB
DynamoDB has no JOIN, no GROUP BY and no table-wide COUNT. How we built a local SQL engine that runs SELECT queries with joins and aggregates over your DynamoDB tables.

· 11 min read

Why a database GUI should stage writes by default
How DynoTable routes DynamoDB edits through a staging area as per-attribute diffs, commits them as transactions guarded by optimistic locks, and gives its AI agent no way to commit.

· 11 min read

What it took to build a safe DynamoDB MCP server
How we built a safe DynamoDB MCP server. An OAuth 2.1 authorization server runs inside the desktop app, each connection binds to one AWS profile and region, and every write waits in staging.

· 15 min read

How we made cheap LLMs reliable at DynamoDB
How we made cheap LLMs reliable at DynamoDB. An error that names the fix, plus a matching prompt rule, took Amazon Nova 2 Lite from failing a join eval to 3 of 3 passes. An audit showed our own evals had scored wrong answers as correct.

More DynamoDB resources

Learn DynamoDB
A structured DynamoDB curriculum, from the fundamentals through to the internals.
DynamoDB Tools
Expression builder, pricing and capacity calculators — right in your browser.

Work with DynamoDB without the Console

A fast DynamoDB desktop client that runs the real SQL DynamoDB can’t — JOINs, GROUP BY, aggregates — with visual editing and an AI agent on your own Bedrock keys.

DownloadSee pricing

Free 30-day trial, no credit card — then the Free plan with no time limit.