Using AI With Distru
For Distru customers and people considering it: get your data out, ask it questions, turn buyer emails into draft orders, and build a small status-update robot.The Distru API in 20 minutes, asking your inventory questions from an export or a read-only connector, AI order intake, and a status-update agent with guardrails.Distru Public API v1, CSV-in-context versus a read-only tool layer, AI order intake, and an agent for order state transitions with guardrails.
Carry on withCarry on withnext:
- 018 min
Your Distru Data, Out of Distru, in 20 MinutesYour Distru Data, Out of Distru: the Public API in 20 MinutesDistru Public API v1: bearer auth, next_page pagination, GET /orders and /packages to CSVDoneDonedone
Every automation in this course starts with getting your data out. Here is the front door, the four facts you need, and how to have AI write the script so you only read it.Every automation in this course starts with getting your data out. The front door is the Distru public API: where the key lives, how pages work, what is rate limited, and a script the AI writes and you check in dry-run before it touches anything.Base URL, Authorization: Bearer, the {data, next_page} envelope, inclusive comma datetime ranges, PDF-only rate limits, strings for money. A stdlib exporter with --dry-run, tested.
- 027 min
Ask Your Inventory a QuestionAsk Your Inventory a Question: the Export Path, the Connector Path, and Which One You NeedInventory Q&A: CSV in context vs. a read-only tool layer over GET /inventoryDoneDonedone
'What is running low, and where?' You can answer that in a chat window this afternoon, from a file you already have. Here is the file way, the live-connection way, and what each one really costs.Attach your export and ask, or give the AI a read-only plug into Distru and let it look. The file way works today and costs nothing; the plug costs tokens on every turn and needs an allowlist. Ten questions to start with, and the one rule that makes answers checkable.Two shapes: CSV in context (cheap, stale, auditable) and a read-only MCP tool layer over the Distru v1 read endpoints (live, costs tool definitions plus rows every turn). GET /inventory requires groupings[], returns ids and decimal strings, omits zero rows. Make the model emit the query, not the number.
- 038 min
From Buyer Email to Draft OrderFrom Buyer Email to Draft Order: Their Last Orders, Your Live Menu, Your ApprovalOrder intake against Distru: parse, match to menu and order history, confidence per line, human write gateDoneDonedone
A buyer texts 'same as last time plus 10 of the new pre-rolls.' Turning that into an order is four jobs, and only one of them is AI. Here is the whole shape, and the ninety-second review that keeps it honest.Parse the buyer's message, resolve it against their last orders and your live menu, put a confidence on every line, and stop. A person approves before anything is written. What Distru's order agent does today, and the version you can build yourself.Two reference sets (GET /orders filtered by company, the live menu) plus per-line confidence, then a human. Distru's order agent populates a cart the rep submits; the DIY path reuses the order-normalizer and ends at POST /orders with no id and status PENDING.
- 049 min
Build a Little Robot That Updates Order StatusBuild a Small Agent That Updates Order Status, and Put Rails Around ItState-transition agent: allowlist, find-before-write, idempotent ledger, stopping condition, dry runDoneDonedone
'Can it just set today's deliveries to Delivering every morning?' Yes. The five-minute part is the work; the rest of this lesson is the rails that keep it from touching the wrong order.A real support request, built end to end: find today's orders, check the move is allowed, re-read before writing, write once, log everything, stop at a limit. Almost none of it is AI, and the parts that are get a gut check first.GET /orders by delivery_datetime and status, an ALLOWED_TRANSITIONS table, re-read by id before each POST, a jsonl ledger for idempotency, MAX_WRITES_PER_RUN, an optional decision-model gate, --dry-run against a fixture.