🔍 Read the full analysis: Why I Set Up A Shared Workspace For My AI Agents on ThorstenMeyerAI.com
TL;DR
Thorsten Meyer says he is building Thrymvault as a shared workspace for AI agents in his network of news sites, after chat and a code-focused task tracker made editorial work hard to oversee. The system is designed to record task status, editor instructions and which approved application made each change; backup hardware and other elements are still being rolled out.
Thorsten Meyer says he is setting up Thrymvault, a self-hosted workspace where editors and AI agents can track newsroom tasks, drafts and decisions, after chat and a code-oriented task tracker proved difficult to manage. The project is intended to make agent work more visible and changes easier to trace across StrongMocha News Group, his network of niche and news sites.
Meyer described the change in a November 13 article on ThorstenMeyerAI.com. He said that on one day that month, his team steered nine articles through a chat window and a task tracker built for software work. Agents gathered sources, drafted copy and prepared graphics, but he had to ask for status updates and issue headline changes in chat, relying on the right agent to notice and act.
Thrymvault is presented as a shared place for pages, databases and boards, hosted on Meyer’s own machines. Agents are intended to access it through approved applications using an MCP server with 20 tools, a REST API or a command-line tool. Meyer says the agent guide is generated from the same tool catalogue as the server, so its instructions match the available tools.
In the proposed editorial workflow, a story begins as a card on a board. Moving it into a “Kit” stage prompts an editorial agent to gather sources and check figures; the agent then drafts the article and updates a report linked to the card. Editors can request revisions in the card’s comment thread by mentioning the relevant application. The workflow ends with a published link, while a “stopped” card tells agents to leave the item alone. Meyer says the board movement is an agreement for agents to follow, not a software switch that automatically starts work.
The shared desk · animated
How a story moves through the shared desk
One card per story. The editor moves it and asks questions in its thread; the agent does the work, keeps its status current and answers there. Every agent change is signed with its application.
Editorial board
Thread on the card
Editor@Editorial agent shorter title, no colon
AgentDone: title shortened, draft updated.
via Editorial agent
Signed history
- Moved to Kit · by the editor
- Status and draft link updated · via Editorial agent
- Comment mentions Editorial agent · by the editor
- Reply in the same thread · via Editorial agent
- Approved and scheduled · by the editor
- Published link added · via Editorial agent
Agents add and edit. They cannot delete anything.
Source: Thrymvault agent integration guide and rollout plan, November 2026. Shows the desk as designed for StrongMocha News Group; the board templates and the pick-up of mentions are still being rolled out.
Animation: Thorsten Meyer AI
A More Traceable Editorial Workflow
The shift matters because AI agents can work on several newsroom tasks at once, while a stream of chat messages can leave editors unsure about what is finished, which instruction is current or who changed a draft. A shared board gives the editor a view of the work and keeps comments and reports attached to the relevant story rather than scattered through conversations or stored on individual machines.
Meyer also frames the workspace as a way to make corrections and oversight more practical. He says changes are marked with the approved application that made them, and decisions such as a new headline rule can be recorded with a date, scope and link to the rule. That could help a newsroom review how a piece was produced. It does not, by itself, establish that an agent’s sources or copy are accurate; editorial review remains part of the described process.
For other organizations experimenting with agents, the account offers a specific operational argument: instructions and permissions should live in a shared, auditable workspace rather than depend on a chat history. The system is Meyer’s own newsroom project, however, and the material provided does not include independent testing or evidence that it improves speed, accuracy or cost.
Popular right now
Shark WANDVAC Power PET Cordless Handheld Vacuum, White, WV410WH
Ninja Foodi DualZone Smart XL Air Fryer, 10 Qt, 6-in-1, Black (DZ550)
De'Longhi Classic Signature Espresso Machine with Adjustable Milk Frother
As an affiliate, we earn on qualifying purchases.
From Chat to a Newsroom Desk
Meyer says the immediate impetus came from coordinating multiple articles through a chat interface and a tracker intended for code work. In his account, story status was not readily visible, requests could scroll out of view, and finished explainer pages were not easy to share as links. Those are his descriptions of the existing workflow, not findings from an outside assessment.
He compares Thrymvault’s general shape to Notion, describing it as a workspace for pages, databases and boards. Its stated distinction is that it runs on the group’s own hardware and gives agents limited access through approved applications. Meyer says each application has a key tied to the rights of the person who created it and is confined to one workspace. He also says agents cannot delete content. These are design and product claims drawn from the site’s guide and development notes, as cited in his article.
The article separates the intended workflow from what is already operating. It describes some agent tools and access rules as built, while identifying a second machine on standby and hourly backups as a next step rather than a live feature. The newsroom process is therefore a mix of current functionality and planned rollout, not a claim that every element is already in routine use.
„That is the point where an AI-run newsroom outgrows chat.“
— Thorsten Meyer, writing on ThorstenMeyerAI.com
Rollout and Safeguards Still Developing
The article does not specify how many Thrymvault features are already in daily newsroom use, how reliably agents respond to card changes or mentions, or what review process checks their reporting before publication. It also does not provide independent security testing, performance data or comparisons with other newsroom systems.
Meyer identifies a standby machine and hourly backups as planned work, not a current capability. The article does not give a completion date for those steps. It also leaves unclear how access would be revoked, how long change records are retained, or how the workflow handles disagreements between an agent’s output and an editor’s instructions.
Backup Hardware and Wider Rollout
Meyer says the next infrastructure step is a second machine on standby with hourly backups. He does not give a date for putting that arrangement into service. The article also signals that parts of the agent interface and newsroom process are still being rolled out, but does not lay out a release schedule or adoption milestones.
The next evidence readers can look for is whether the described workflow becomes routine: whether story cards, source kits, draft reports and revision threads are consistently used, and whether the planned backup system goes live. Meyer has not published metrics showing the system’s effects on newsroom output, accuracy or editorial workload.
Key Questions
What is Thrymvault?
Thrymvault is a workspace for pages, databases and boards that Meyer says his group builds and hosts on its own machines. It is intended to give editors and AI agents a shared place to manage newsroom work.
Why does Meyer say chat is not enough?
Meyer says chat makes story status and instructions hard to track as messages accumulate. He wants tasks, editor comments and agent reports attached to a shared story card instead.
How are agents meant to access the workspace?
The article lists an MCP server with 20 tools, a REST API and a command-line tool. Meyer says agents work through applications approved by the workspace owner and that changes identify the application involved.
Are all of the safeguards and infrastructure live?
No. Meyer says agents cannot delete content and describes approval-based access and attributed changes in the current design. A second machine on standby and hourly backups are described as planned, not live.
Does the article show that Thrymvault improves accuracy?
No performance or accuracy results are provided. Meyer argues that a traceable workflow can make errors easier to investigate and correct, but the article does not report independent testing or measured outcomes.
Source: ThorstenMeyerAI.com
