cd /news/ai-agents/a-preview-of-mefis-studio-ai-0-4-age… · home › topics › ai-agents › article
[ARTICLE · art-139751] src=dev.to ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

A preview of Mefi’s Studio AI+ 0.4: Agent Brain, Project Map, and coordinated agents

A developer is previewing version 0.4 of Mefi's Studio AI+, a desktop workspace for coding with AI agents, which introduces an Agent Brain live pipeline view with replayable work events, a Project Map built from Git history and agent-read files, and coordinated sub-agents with optional nested delegation up to three levels. The unreleased update also adds a Playbook for reusing task recipes, ask_desk support for OpenCode and Claude Code, OpenRouter as a companion provider, and reorganized Home, Work, Live, and Models navigation. The developer is soliciting feedback from coding-agent users on what to surface when multiple workers run and whether activity replay or a project map is more useful on returning to a project.

by read2 min views2 publishedSep 25, 2026

I'm working on 0.4 of Mefi's Studio AI+, a desktop workspace for coding with AI agents. The focus of this update is making it easier to see what agents are doing, how their work connects, and where they need help.

This is a development preview. The package is still at 0.3.3, and the 0.4 work is listed as Unreleased, so the final release scope may change.

Agent Brain is the centerpiece of the current 0.4 candidate. Its Live view shows a task's pipeline, lead agent, and sub-agents. Work events can be replayed so you can look back through a day's activity. Pipelines can grow from worker todos and MEFI_STEP updates, with completed steps folded away.

Project Map builds a view of project systems from Git history and the files agents read or change. It connects related systems and brings relevant files into task briefs. The map sits beside the conversation, and you can explore a system's tasks, ideas, and plans before starting work in that area.

The coordination work includes sub-agents reporting results to their parent, an optional desk worker for stuck workers, and a Playbook for saving and reusing successful task recipes. Optional nested delegation can go up to three levels. The current work includes ask_desk support for OpenCode and Claude Code, plus OpenRouter as a companion provider.

Navigation is being reorganized around Home, Work, Live, and Models. The companion gets a welcome-back summary and one queue for items that need attention. Plans also gets an AI writing partner that explores project files and offers editable suggestions with evidence.

There are smaller quality-of-life changes too, including clearer project preview controls, live appearance previews, and a combined music and video menu.

I'd like feedback from people who use coding agents: when several workers are running, what do you most need to see to understand whether the task is on track? And when you return to a project, would an activity replay or a project map help more?

If you want to follow the 0.4 work or talk through your workflow, come hang out in our Discord.

── more in #ai-agents 4 stories · sorted by recency
── more on @mefi 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/a-preview-of-mefis-s…] indexed:0 read:2min 2026-09-25 · —