TL;DR
Thorsten Meyer AI says Threlmark is built as a Next.js app over plain JSON files, making the on-disk layout the application contract. The design is intended to let users inspect, sync and version project data while allowing external tools and AI agents to update cards without a cloud service or account system. Several details remain unclear, including release timing, multi-device conflict behavior and production hosting plans.
Thorsten Meyer AI has published a technical account of Threlmark’s local-first architecture, saying the project tool uses plain JSON files on disk rather than a database, cloud service or user accounts. The report matters because it describes a project-management system built around interoperability with external tools and AI coding agents, with the file layout itself treated as the API.
Confirmed from the provided source material: Threlmark is described as a Next.js and TypeScript app whose data root defaults to ~/.threlmark. The layout includes a manifest, dependency graph, project metadata, board ordering, item files, suggestions, handoffs, reports, a roadmap mirror, shared items and an archive.
The report says each card is stored as its own JSON file, while lane order lives separately in board.json. According to Thorsten Meyer AI, that choice is meant to reduce write conflicts because external tools can add or edit item files without touching the board ordering file. The board then reconciles items when Threlmark reads the project data.
The source also says Threlmark uses atomic writes: data is written to a temporary file in the same directory, synced, and then renamed over the target file. The stated goal is to avoid partially written JSON files after a crash. Metrics such as priority, work-item age, cycle time, throughput and work-in-progress counts are described as computed from item state rather than stored as separate values.
Why It Matters
The architecture is relevant to readers following local-first software, developer tools and AI-assisted work because it treats ordinary files as the system boundary. If the model works as described, users can inspect, diff, back up, sync or version project data with existing tools instead of relying on a hosted database.
The agent workflow is also central to the report. Threlmark is described as allowing an AI coding agent to receive a handoff, report completion through REST or a filesystem drop, and move a card to Done. That design links task planning, implementation status and agent reporting in a way that could reduce manual project tracking for solo developers and small teams.
JSON file editor for project management
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background
Threlmark is framed in the source as a project tool that asks what the most important next task is across a user’s work, and whether an assigned AI agent actually shipped it. The report contrasts that goal with simpler board tools that mainly track which column a card is in.
The source places this architecture article as part 2 of a Threlmark series. It also describes a portfolio score that ranks items across projects using priority, status weighting and blocker count. The provided material mentions a static read-only demo using seeded data and localStorage, but the excerpt cuts off before fully explaining the personal Node deployment path.
“the on-disk layout IS the API”
— Thorsten Meyer AI report
“There is no server-of-record”
— Thorsten Meyer AI report
“Derived, never stored”
— Thorsten Meyer AI report
“A handoff is a first-class flow event”
— Thorsten Meyer AI report
atomic file write software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Remains Unclear
The source material does not give a publication date, release status, repository link, security model, license confirmation beyond an MIT reference, or full deployment instructions. It is also unclear how Threlmark handles simultaneous edits to the same item file across synced folders or network drives. The excerpt states that unknown keys are preserved, but it does not show compatibility tests or examples from third-party tools.
local-first project management tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What’s Next
The next developments to watch are a public release, install instructions, repository documentation, or further posts in the series that fill in deployment and sync behavior. For now, the report establishes the design claims; independent verification would require access to the code, test results or a working build.
file-based project tracking software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the main development?
Thorsten Meyer AI published a technical report explaining how Threlmark uses JSON files on disk as its application contract.
Does Threlmark use a database?
According to the source material, Threlmark does not use a database, cloud service or user accounts. It stores project data in plain files under a local data root.
How can AI agents update cards?
The report says agents can report completion through a REST endpoint or by dropping a report file into the project reports folder. A done report can move the related card to Done.
What remains unknown?
The material does not fully answer release timing, production hosting, security controls, third-party adoption or how conflicts are resolved when the same item file changes in more than one place.
Source: Thorsten Meyer AI
