TL;DR
Thorsten Meyer AI has published a technical account of Threlmark, a local-first project tool that uses plain JSON files on disk instead of a database, cloud service or user accounts. The report says the file layout acts as the API, allowing the UI, outside tools and AI agents to work against the same records.
Thorsten Meyer AI has published a technical account of Threlmark, describing the project tool as a Next.js and TypeScript app that stores its records as plain JSON files on a user’s disk rather than in a database or cloud account, a design choice that affects how the product handles task state, external tools and AI agent handoffs.
The report says Threlmark’s data root defaults to ~/.threlmark, with a manifest, dependency graph, project folders, one JSON file per item, suggestion inboxes, handoff records, reports and a human-readable roadmap file. In the architecture described by Thorsten Meyer AI, the file system is not a backup layer or export format; it is the record that both the application and outside tools read and write.
According to the source material, Threlmark avoids a single server-of-record and does not require a database, cloud service or user account. The article says this makes project data inspectable through normal file tools, portable through copy, sync or version control, and accessible to tools written in other languages.
The report also describes two safety patterns used to make file-backed state less fragile: atomic writes and one file per card. Writes are made to a temporary file in the same directory before being renamed over the target file, while board order is kept separately in board.json and reconciled when read. The source says unknown keys are preserved so other tools can add data without breaking the contract.
Why It Matters
The architecture matters because it reflects a growing local-first approach to software used with AI coding agents. If the source material is accurate, Threlmark treats the disk as a shared interface among the user interface, automation tools and agent workflows, rather than placing all coordination behind a hosted API.
That could make the tool easier to inspect, back up, diff and commit, especially for developers who already manage work through files and version control. It also lowers the barrier for other local tools to participate, since they can create or update item files without negotiating account permissions or writing against a remote database.
The agent workflow is the most direct product implication. Thorsten Meyer AI says a card can be handed to an AI agent, receive a report through a REST endpoint or a filesystem drop, and move itself to Done when a completion report is accepted. That links planning, execution and status updates in one file-based loop.
Python in Action: 60 Mini Projects to Automate Everything (Part 1): Practical CLI Tools, File Automation, and Data Cleaning with CSV, Excel, and JSON
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background
The report positions Threlmark as a project tool focused less on which column a card occupies and more on identifying the next highest-value task across active work. It says priority is computed on read from factors including impact, evidence, fit and effort, rather than being stored as a fixed field that could drift from the underlying item data.
Other metrics described in the source material are also derived from the item files. Age, cycle time, throughput and work-in-progress counts are calculated from item state and append-only lane changes. Portfolio ranking is described as status-weighted, with in-flight items and blocked work rising in the list based on formula-driven scores.
The supplied material also mentions deployment paths, including a static read-only demo using seeded data and localStorage, plus a personal Node path, though the latter description is cut off in the provided source.
“the on-disk layout is the API”
— Thorsten Meyer AI
“There is no server-of-record — the files are the record”
— Thorsten Meyer AI
“A handoff is a first-class flow event”
— Thorsten Meyer AI
The Phoenix Project: A Novel about IT, DevOps, and Helping Your Business Win
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What Remains Unclear
The supplied source material does not provide an exact publication date, independent benchmark results, production adoption figures or a full security model. It is also not clear from the excerpt how Threlmark handles simultaneous writes across network-synced folders, access control on shared machines, schema migration over time or recovery from malformed third-party item files.
The personal Node deployment path is referenced but truncated in the provided material, so the full hosting guidance and write-mode limitations are not confirmed here.
file-based task management app
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 project publishes fuller implementation details, release notes, code-level documentation, or tests showing how the file contract behaves under concurrent edits and agent-generated reports. Adoption will depend in part on whether developers trust the file layout as a stable integration surface.
Next.js JSON project tool
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 published a technical report describing Threlmark’s local-first architecture, centered on JSON files stored on disk as the system’s main record and integration contract.
Is Threlmark cloud-based?
According to the source material, Threlmark is described as a Next.js app with no database, no cloud service and no user accounts. Its data root defaults to ~/.threlmark.
How does Threlmark avoid corrupting files?
The report says it uses atomic writes, writing first to a temporary file and then renaming it over the target, and stores one card per file to reduce write collisions.
How do AI agents fit into the system?
Thorsten Meyer AI says Threlmark can hand a task to an AI agent, receive a report by REST or through a reports folder, and move the card to Done when a completion report is accepted.
What remains unclear?
The provided source does not confirm release timing, adoption, benchmarks, full deployment details, or how the system handles every edge case around sync conflicts, permissions and malformed external writes.
Source: Thorsten Meyer AI
