TL;DR
Thorsten Meyer AI published a technical report describing Threlmark as a local-first project tool built on plain JSON files rather than a database, cloud service or account system. The report says the file layout itself acts as the application contract, allowing the UI, external tools and AI coding agents to work against the same on-disk data.
Thorsten Meyer AI has published a technical report describing Threlmark as a Next.js project tool that runs on plain JSON files stored on a user’s disk, a design the report says avoids a database, cloud backend and user accounts while allowing AI coding agents and outside tools to update project work directly.
The report says Threlmark’s default data root is ~/.threlmark, with a manifest, dependency graph, project folders, per-project board data, one JSON file per card, handoff records, report drop zones and a human-readable roadmap mirror. According to the source material, the application treats that on-disk structure as the API shared by the UI and external tools.
The system is described as using atomic writes: data is written to a temporary file in the same directory, synced, then renamed over the target. The report says this pattern is meant to leave either the old complete file or the new complete file after a crash, rather than a partial write. It also says Threlmark stores each work item in its own file to reduce write collisions when more than one tool is active.
Several values are derived from files rather than stored directly, according to the report. Priority is calculated from impact, evidence, fit and effort; staleness is based on how long an item has remained in a lane; cycle time is derived from state-change history; and throughput is counted from items reaching Done. The source says this is meant to keep displayed metrics aligned with the underlying JSON records.
Why It Matters
The report matters because it points to a different path for project software at a time when more work is being passed to AI coding agents. Instead of making a hosted service the system of record, Threlmark’s design makes the user’s local files the record, which may appeal to developers who want inspectable data, easy backup, version control and tool interoperability.
The architecture also changes how automation can participate. According to the report, an outside tool can create or update an item by writing a file, while Threlmark reconciles board state on read. That lowers the integration burden for local tools, scripts and agents, though the source material does not provide independent performance, reliability or adoption data.
plain JSON file project management tool
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background
Threlmark is presented as a project-management tool aimed at ranking work across projects and tracking whether AI-agent handoffs are completed. The report contrasts that with tools centered on board columns, saying Threlmark focuses on identifying the next highest-priority task across a portfolio and closing the loop when an agent reports the work as done.
The source material says cards can live under a project’s items directory, while shared cards can be referenced across projects. Lane order is stored separately in board.json. The report says the board can repair itself when it reads item files, which is why external tools do not need to edit the board-order file directly.
The report also describes two deployment paths: a static read-only demo using seeded data and localStorage, and a personal Node-based setup for local use. The provided material is cut off before all hosting details are available.
“the on-disk layout is the API”
— Thorsten Meyer AI report
“There is no server-of-record”
— Thorsten Meyer AI report
“A handoff is a first-class flow event”
— Thorsten Meyer AI report
local-first project management 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 state whether Threlmark has shipped as a stable release, how many users or projects are using it, or what failure modes have been tested beyond the stated file-write patterns. It is also unclear from the provided material how Threlmark handles cross-device sync conflicts, permissions on shared machines, corrupted JSON files, or large project sets.
version control JSON project files
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What’s Next
The next items to watch are whether the full Threlmark project, documentation and code examples are published, whether the local Node setup is detailed, and whether the system’s agent-reporting workflow is tested across common AI coding tools. More evidence would be needed to judge how the architecture performs outside the described design.
AI coding agent project tools
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 Threlmark’s local-first architecture, centered on JSON files stored on disk as the system of record.
Does Threlmark use a database or cloud backend?
According to the report, no. The described system uses a Next.js app over plain files and does not rely on a database, cloud service or user accounts.
How does Threlmark try to make file writes safer?
The report says Threlmark uses atomic writes and stores one card per file. That design is meant to reduce partial-write risk and lower the chance that two tools overwrite the same shared array.
How do AI agents fit into the system?
The report says agents can receive handoffs and report completion through REST or by dropping a report file. A completed report can move the related card to Done, according to the source material.
What remains unknown?
The provided material does not confirm release status, user adoption, full hosting details, or how the system handles sync conflicts and other edge cases in everyday use.
Source: Thorsten Meyer AI
