GRUENCY  /  Software  /  WS.AgenticOS

WS.AgenticOS: how GRUENCY keeps its AI work in view.

A local panel over Claude Code. It maps every memory file, skill and routine, shows what each skill and routine costs in tokens, and lists what to fix next.

Runs on our own computerScanner only readsPlain-text memoryNo external libraries

See the memory map Plan one for your team

Concept illustration: two views of a glass android built from workstation cooling parts.
Concept illustration. Map: demo data, not our real files.

What the panel shows

Claude Code does the work. WS.AgenticOS shows what it works with: what it knows, what it can do, when it runs on its own and what that costs.

Memory

Plain-text notes, filed by department. The map shows every file, its links and its age, so stale or orphaned notes stand out.

Skills

Written procedures Claude follows for recurring jobs. The panel counts how often each one runs and how many tokens it uses.

Routines

Scheduled jobs that run without anyone asking. The panel lists each one with its schedule, its weekly cost and a weekly limit.

Usage and next steps

Tokens per day, project, skill and sub-agent. Suggested fixes go into a basket that becomes one instruction for Claude.

What the company knows, on one map

Every memory file is a dot, newer files on the inner rings. Colour marks the area: warm for WS.COMPUTER, cool for GRUENCY. Lines join files that refer to each other, and skills and routines sit outside.

Illustration. Demo data, not our real files. The panel also has a third, force-directed view.

The start screen

It answers three questions: what ran, what it cost, and what needs attention. Ticked suggestions become one instruction for Claude, which a person reads before it runs.

WS.AgenticOS · Start
Memory files320
Skills48
Routines9
Tokens, 7 days40 M

Routines, next 24 hours

  • Price check, WS.COMPUTER06:00done
  • Site health, GRUENCY07:30done
  • Notification digest12:00done
  • Memory review17:00next

This week: 3.1 M of a 6 M token limit

Tokens by department, 7 days

  • WS.COMPUTER22 M
  • GRUENCY12 M
  • System4 M
  • Other2 M

Next steps

  • 12 memory files not opened in 90 days: merge or retirememory
  • 2 skills do mechanical work on the largest model: move to a smaller onecost
  • 1 routine reports even when nothing changed: report only on findingsroutine

2 in basketCopy instruction for Claude

Illustration of the start screen. Demo data, rounded.

How it is built

  1. Memory, skills, routines, session logsPlain files on our own computer
  2. ScannerReads them and changes nothing
  3. PanelAnswers only on this computer
  4. A personPicks what to change
  5. Claude CodeMakes the change, as in any session
Local only
The panel adds no online service. Claude Code itself still talks to Anthropic, as it always does.
Read-only scan
The scanner never edits a file. Every change goes through Claude, with a person approving it.
Plain text
Memory is ordinary text files with an index. No vector database: a person can open, read and correct every note.
In development
Starting skills from the panel in the background, and a calendar of routines. Neither is in use yet.

One system for three kinds of work

Workstations

WS.COMPUTER builds: component prices, product cards, BIOS guides and service reports follow written skills, so each build is documented the same way.

Software

GRUENCY applications and this website: design rules, copy rules and release steps live as skills, and a critic agent scores pages before release.

Research and data

Measurement series, tables and papers: skills hold the method, and the memory keeps what was decided and why.

A system like this for your team

Sebastian Michalski built WS.AgenticOS for GRUENCY in October 2026. Under technical advisory, he can plan a similar set-up with your team: memory, skills, routines and cost limits around your own work.

Read about technical advisory

Concept illustration: close-up of the android torso, with a coolant reservoir, hoses and quick-release fittings.
Concept illustration.