THE DESKARDEN WORKSPACE

A home for your work.
A working home for AI.

Projects give agents context. Workflows give them direction. You keep control of what becomes part of the finished work.

Studio / Projects Local models
PROJECT / PRODUCT LAUNCH

Launch brief

A clear plan, grounded in your project files.

Prepare the launch briefReady for review
  1. ✓Reader · collect source materialComplete
  2. ✓Analyst · compare prioritiesComplete
  3. ✓Editor · prepare the briefComplete
launch-brief.md

Draft ready. Two decisions need your review before sharing.

positioning.pdf · p. 4roadmap.md · §2
Product tour · example project

Explore three example workflows and switch between Overview, Files and Agents. Follow a sample project from source material to review.

01 / PROJECTS

Context belongs
with the work.

A project brings together a brief, source files, instructions, agent conversations and accepted outputs. Keep shared references separate from working drafts so everyone knows what is current.

Source references make an answer inspectable. A passage points back to its document or text section; an unverified finding stays visible as an open question.

02 / AGENTS

Define the role.
Set the boundaries.

An agent combines instructions, a model and permitted tools. Its role might be reading source files, comparing proposals or preparing an update.

Roles can share the same local model. Use a different model when a task needs it; adding a role does not require another server.

Read

Reader

Locate relevant passages, extract facts and identify missing material.

Analyze

Analyst

Compare evidence, surface differences and organize a recommendation.

Draft

Editor

Turn findings into a useful document with the right structure and tone.

Review

You

Resolve uncertainty, approve changes and choose what gets published.

03 / RUNS & REVIEW

Know what happened.
Decide what happens next.

A run records its instructions, selected sources, steps and outputs. Independent steps can work in parallel within your host’s capacity; dependent steps wait for the material they need.

Review is part of the workflow. A draft can be ready while a proposed file change or outbound action is still waiting for permission.

Understand run states →
04 / MODELS & CAPACITY

Your model choices.
Your resource budget.

Match the task to the model and the amount of work your host can sustain.

Local inference

Choose an open-weight model served by a local runtime. Model licensing and runtime support depend on the model you select.

Prepared context

Prepare source material once and reuse it across related tasks. Keep references tied to the version of the source used.

Bounded parallel work

Set worker limits to fit your memory and compute. More simultaneous work is useful only while the host has room for it.

See what a workflow can do.

Start with a concrete task and follow it from source material to a reviewed output.

Explore workflows ↗