TL;DR
Thorsten Meyer AI has published a technical report describing Threlmark, a local-first project management tool built on plain JSON files instead of a database, cloud service or user accounts. The report says Threlmark treats its on-disk folder layout as the API, allowing the UI, external tools and AI coding agents to work from the same files.
Thorsten Meyer AI has published a technical report on Threlmark, a local-first project management tool that runs as a Next.js app over plain JSON files rather than a database, cloud service or user accounts, a design aimed at making project state readable, portable and usable by AI coding agents.
The report says Threlmark’s core architectural choice is that its on-disk layout serves as the application interface. By default, the system stores data under ~/.threlmark, with files for the manifest, dependency graph, project metadata, board ordering, individual work items, agent handoffs, reports and a human-readable roadmap mirror.
According to the source material, each card is stored as a separate JSON file under a project’s items directory. Lane order is kept separately in board.json, and the board reconciles item state when it reads from disk. The report says this pattern allows external tools to add items without editing board ordering directly.
The report also describes two safety patterns for file-based state: atomic writes and one file per item. Atomic writes use a temporary file in the same directory followed by a rename over the target. The report says this avoids partially written files after a crash, while one-file-per-card storage reduces write conflicts between tools.
Why It Matters
The architecture matters because it treats local files as a shared operating surface for humans, apps and AI agents. If the design works as described, a user can inspect, diff, back up, sync or migrate project data with ordinary file tools instead of relying on a hosted account or proprietary database.
The report also points to a broader shift in software built around AI coding agents. Threlmark is described as tracking not only the status of work items but whether an assigned agent completed the work and reported back. A completed report can move a card to Done through either a REST endpoint or a filesystem drop zone, according to the source material.
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
Threlmark is presented as a project tool focused on ranking what should be done next across multiple projects, rather than only showing which lane a card sits in. The source material says priority is calculated when data is read, using impact, evidence, fit and effort, instead of being stored as a fixed value in the item file.
The report also describes portfolio ranking across projects using status-weighted scores. Items in development are weighted higher than ranked items, while completed items receive low weight. Blocked dependencies can raise an item’s score, according to the formula described in the report.
The source material frames Threlmark as MIT-licensed and built with Next.js, TypeScript and JSON-on-disk storage. It also describes two deployment paths: a static read-only demo using seeded data and localStorage, and a personal Node setup for writable local use.
“the on-disk layout IS the API”
— Thorsten Meyer AI technical report
“There is no server-of-record”
— Thorsten Meyer AI technical report
“The board heals itself”
— Thorsten Meyer AI technical report
“A handoff is a first-class flow event”
— Thorsten Meyer AI technical report
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 source material describes Threlmark’s architecture but does not provide independent benchmarks, user adoption data, failure-rate testing or a dated release milestone. It is also not clear from the provided material how the system handles multi-device conflicts, permission boundaries, malformed external writes or long-running sync conflicts across Dropbox, Git or other file-sync tools.
Practical Python Projects Step-by-Step: Build Real Applications, Automation Tools, and Desktop Programs with Python 3.14 (Modern Python Development Mastery Series)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What’s Next
The next points to watch are whether Threlmark’s local writable setup gains wider testing, whether external tools adopt the file contract, and whether the agent handoff and report flow proves reliable outside controlled use. More detail would also be needed on packaging, installation, sync behavior and production limits for larger project collections.
Next.js 16.2.2 Unlocked: Advanced Development Patterns for High-Performance Apps
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the actual news development?
Thorsten Meyer AI published a technical report explaining Threlmark’s local-first architecture, including its JSON-on-disk data model, file contract and AI agent handoff flow.
Is Threlmark using a database?
According to the report, no. Threlmark stores project data in plain JSON files on disk, with one file per card and separate files for board order, project metadata, links, reports and handoffs.
What is confirmed and what is claimed?
The confirmed development is the publication of the architecture report. The technical behavior described, including atomic writes, board reconciliation and agent-driven card movement, comes from Thorsten Meyer AI’s source material and was not independently tested here.
Why would readers care?
The design offers a different path for project tools: local ownership of data, file-level portability and a direct way for AI coding agents to report completed work. That could matter to developers who want project state that is scriptable, inspectable and not tied to a hosted account.
What remains unclear?
The provided material does not say how Threlmark performs at scale, how it resolves sync conflicts across devices, or how it prevents accidental or unsafe writes from external tools.
Source: Thorsten Meyer AI
