TL;DR
Thorsten Meyer AI has detailed Threlmark’s local-first architecture, describing a Next.js project tool that treats plain JSON files on disk as the system contract. The report says the design avoids a database, cloud service and user accounts, while allowing outside tools and AI agents to read, write and report against the same file structure.
Thorsten Meyer AI has published a technical report on Threlmark, a local-first project management tool built as a Next.js app over plain JSON files, a design that matters because it makes the file system, rather than a database or cloud service, the record for project work.
The report says Threlmark does not use a server of record, cloud account system or database. Instead, its data root defaults to ~/.threlmark, where a manifest, dependency graph, project folders, item files, suggestions, handoffs, reports and an archive live as ordinary files.
According to the source material, Threlmark uses one JSON file per card. Project metadata is stored separately from board ordering, while individual items live under a project’s items directory. The report says this lets outside tools write a card file without directly changing the board ordering file; Threlmark then reconciles the board state when it reads the files.
The architecture also includes atomic writes. The report describes a pattern in which the app writes to a temporary file in the same directory, syncs it, and then renames it over the target file. The stated aim is to avoid half-written JSON after a crash, leaving either the old complete file or the new complete file.
Why It Matters
The report frames Threlmark as a project tool designed for users who want project data to remain inspectable, portable and usable by other programs. Because each artifact is a file, users can inspect it with command-line tools, back it up with ordinary file copying, sync it through services such as Dropbox or Git, and let tools written in other languages take part by reading and writing the same structure.
The agent workflow is the other main point. The source material says Threlmark can hand off a ranked item to an AI coding agent, accept a report through a REST endpoint or by reading a report file, and move the card to Done when the report marks the work complete. If that works as described, the tool is built around a closed loop between planning, implementation reporting and board state, rather than treating AI work as an external note.
Real-World Android App Projects with Kotlin and Jetpack Compose: Build Production-Style Android Apps with Modern Architecture, API Integration, State Management, Local Data Storage, Practical Projects
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background
Threlmark is presented in the source material as part two of a series and as a technical report on the project’s architecture. The report contrasts Threlmark with project tools that mainly answer where a card sits on a board, saying Threlmark is aimed at ranking what should happen next across work and checking whether delegated agent work shipped.
The source also says Threlmark derives metrics rather than storing them when they can be computed from item state. Priority is calculated on read from impact, evidence, fit and effort. Other derived measures named in the material include stale age, cycle time, throughput and work-in-progress counts.
For portfolio ranking, the report says items are globally addressable and scored with status weights, so work in development, ranked work, ideas and completed items are ranked differently. Blocked items receive an added score bump tied to blocked count and priority, according to the source material.
“the on-disk layout is the API”
— Thorsten Meyer AI report
“There is no server-of-record — the files are the record”
— Thorsten Meyer AI report
“Anything computable from item state is computed”
— Thorsten Meyer AI report
Dinceer 12 Pocket Spiral Project Organizer Folder, Morandi Color Multi Pocket Folder with Inner Zipper Pouch for School, Office, Bills and Documents
POCKET PROJECT ORGANIZER :This spiral project organizer folder keeps documents separated and easy to find, making it a…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Remains Unclear
The source material does not state how widely Threlmark is used, whether the architecture has been tested under heavy multi-user or network-synced workloads, or what failure cases remain when several tools edit the same project at nearly the same time. It also does not provide a release date, version number or independent benchmark data.
Mastering PowerShell Scripting: Automate repetitive tasks and simplify complex administrative tasks using PowerShell
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What’s Next
The next step for readers is to watch whether Threlmark’s file contract holds up in real use across multiple tools, sync systems and AI agents. Future updates would need to clarify packaging, deployment options, conflict behavior and the maturity of the REST and filesystem report paths.
Introduction to AI Agents: A Beginner-Friendly Guide to Building Useful Assistants with Tools, Memory, and Workflows
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the actual development?
Thorsten Meyer AI has published a technical report describing how Threlmark stores project work in plain JSON files and treats the on-disk layout as the interface for the app and outside tools.
Does Threlmark use a database?
According to the source material, no. The report says Threlmark is a Next.js app using files under ~/.threlmark as the record, with one file per card and separate files for project metadata, board ordering and related data.
How does it handle AI agent work?
The report says a card can be handed off to an AI agent with a reporting protocol. The agent can report completion through REST or by dropping a report file, and a completed report can move the card to Done.
What remains unconfirmed?
The source does not provide usage numbers, independent test results or a detailed account of conflict handling across all sync tools. Those points remain open based on the supplied material.
Source: Thorsten Meyer AI
