Work
What we've put into production.
Alacrity is a small studio, so the honest measure is what has actually shipped — including products we built and run ourselves. Four we can talk about openly are below: the problem, the system, the outcome, and what building each one taught us.
In production
- BuildMyLease
- Textral
- kbRelay
- DJKMD Presents Legends
Built and run by Alacrity
BuildMyLease
Statute-grounded lease generation — AI-built legal content, paralegal-validated.
- Problem
- A small landlord needs a lease that is actually correct for their state — the right deposit limits, the required disclosures — without buying property-management software or trusting a random Word template. Get one statutory detail wrong and the document is worse than none.
- System
- BuildMyLease is a focused product Alacrity designed, built, and operates. A landlord picks their state and document type, answers a guided questionnaire, pays once, and downloads a lease that reflects both their answers and their state's statutory requirements — no account, no subscription. The hard part is the legal content itself: the per-state compliance matrices, statutory clauses, disclosures, and editor content were built agentically — AI-assisted research and integration — and then signed off by a paralegal before any of it reached a customer. E-signature is built in to an E-SIGN / UETA-compliant standard.
- Outcome
- A live product at buildmylease.com, paid for one lease at a time, operating across 20 US states for both leases and month-to-month agreements. It runs largely on its own: a narrow, trustworthy flow that is found through search and finished in one sitting.
What building it taught us — It is the clearest version of a principle the whole studio runs on. AI did the heavy lifting — the legal research, the per-state matrices, the assembly — and a human with the right expertise, a paralegal, validated it before it shipped. On legally consequential work, that pairing is the product: AI for reach and speed, human judgment on the line where a mistake would matter.
Built and run by Alacrity
Textral
Our own citation-grounded RAG, in private beta — and we run on it.
- Problem
- Retrieval systems fail quietly. An answer that has dropped its sources, or whose quality has drifted after an embedding change, looks fine right up until someone downstream trusts it and is wrong.
- System
- Textral is Alacrity's own citation-grounded retrieval platform, in private beta with real users. Every answer carries its sources, and when it cannot ground a claim it says so — full, partial, or cannot-answer — instead of inventing one. It is multi-tenant down to the storage boundary, MCP-native so agents can drive ingest and retrieval directly, and it locks each namespace to its embedding model so a silent mismatch surfaces as an error rather than as quietly worse results.
- Outcome
- The same citation-grounded retrieval we would build into a client system, running as a product we own and operate in beta. Dogfooding it keeps us honest about what production retrieval actually costs to run — and it means the RAG we recommend is the RAG we run ourselves.
What building it taught us — Audit-first retrieval earns its keep. Recording every query's grounding and provenance from the start is what turns “the answers got worse and nobody knows why” into an alert you can act on the day it happens.
Built and run by Alacrity
kbRelay
A kanban board where humans and agents relay work to each other — API-first, provenance on every move.
- Problem
- AI agents are starting to do real work, but the tools we track work in still assume a person behind every action. When an agent files a task, moves a card, or leaves a note, there is no honest record of who — or what — did it, and no clean way for a person and an agent to hand a task back and forth without stepping on each other.
- System
- kbRelay (Kanban Relay) is a Trello-style board Alacrity built for a world where agents work alongside people. Every card records who created and last touched it, and whether they were a human or an agent, so the board is an honest ledger of agentic work. It is API-first — everything the web board does, the HTTP API does, enforced by a parity test — and agents connect through a published MCP server with one command. Admins mint dedicated agent identities with their own keys, so an agent's work is attributed to it, never a borrowed human login. A card's spec and its append-only timeline are kept separate: edit the plan in place, report results on the log, never rewrite history. The same codebase runs on Cloudflare (Worker + D1 + Pages) or self-hosted on a single Docker container.
- Outcome
- A live product at kbrelay.lalalimited.com with a published MCP server on npm — and the system Alacrity runs its own agent work on. The handback contract behind this very site — ready → in progress → in review, a human closes it — is a kbRelay board.
What building it taught us — For human–agent collaboration, provenance and a clean contract are the product. Making “who did what, and were they a person or a machine” a first-class fact — and keeping the spec separate from the log — is what lets people trust and audit work that agents increasingly carry. Same principle as the rest of the studio: the value lives in the boundary.
Client
DJKMD Presents Legends
Agentically-controlled show management, driven over MCP.
- Problem
- Legends is a New England live-performance group whose business is filling rooms — scheduling shows, selling tickets, fielding booking inquiries, and checking guests in at the door. That work was scattered across tools that did not talk to each other.
- System
- Alacrity built the operation as a documented interface first. Shows, ticketing and payment (Square), bookings, and door check-in all sit behind one API with a machine-readable contract and runbooks written for an agent to follow — exposed over MCP so a coding agent can drive show management directly. It can add a show, update a lineup, or mark a night sold out on the live site in about a minute — no redeploy, no engineer in the loop. Capacity is enforced automatically: when a show fills, a payment webhook flips it to sold out.
- Outcome
- A highly profitable, growing live-performance business — tickets sold, rooms filled — where day-to-day show management is driven agentically against a stable contract rather than by a person editing code.
What building it taught us — An agentic system is only as good as the contract beneath it. Writing the API and its runbooks so a machine can operate them safely — explicit, documented, hard to misuse — is what makes hands-off, MCP-driven operation trustworthy. Most of the value lives in the boundary, not the model.
The common thread
Every one of these ran the same way.
Different domains, one method: diagnose the real workflow, build the system around it, and run it in production with the boundaries and measurement that keep it honest.
Start here
Tell us about the workflow.
A few sentences is enough. We'll reply within a business day with an honest read: whether AI creates leverage there, and what it would take to ship.
Replies come from [email protected]