Agent spaces
for your business.
Pick a space — sales, marketing, video, business data — and get it running for your business, on its own or as part of Vibemind. Each space says whether its code is open source.
Open source · self-hosted · no telemetry in the runtime · join the Vibemind beta
Spaces, honestly labelled.
Each space works on its own or as part of Vibemind. Live runs on our own server today; in progress exists in code but is not finished; research is not offered yet.
Live · 5
Sales Claw
liveCompany research, CRM records and outreach drafts across several channels.
Every outgoing message needs human approval
Marketing Agent
liveNewsletters, campaigns and images, approved by a person before anything goes out.
Generation runs on a workstation
Open source · MIT (part of vibemind-os)
Business Data Layer (Rowboat)
liveA knowledge vault with semantic search across everything you capture.
Open source · Apache-2.0 · fork of rowboatlabs/rowboat
In progress · 11
Learning (Academy)
in progressCourses, exercises and a tutor built from your own material, on LearnHouse.
Invitation only; licence file pending (AGPL-3.0)
Personal Space
in progressReads how your day went and helps you choose what comes next: creative work, structured work, or a break when you are close to burning out.
Not running today; code not in the public repository yet
Ideas
in progressCapture and link ideas, then promote them into a project.
Runs on a workstation only
Open source · MIT (part of vibemind-os)
Desktop Automation
in progressClick, type and take screenshots on your own machine.
Workstation only
Open source · MIT
Deep Research
in progressStructured web research.
Part of the OpenFang runtime
Open source · Apache-2.0 OR MIT · fork of RightNow-AI/openfang
Scheduling
in progressReminders, timers and planned tasks.
Off by default
Open source · MIT (part of vibemind-os)
Requirements Factory
in progressCapture requirements in a structured way.
Not running today
Open source · MIT (part of vibemind-os)
Video from August 2026
Software Design
in progressArchitecture work before the code is written.
Not running today
Video from August 2026
Coding
in progressDescribe an app and get a generated project.
Engine not running
Open source · Apache-2.0
Project Intake
in progressA Telegram assistant works through a question tree with you and writes a project brief as a PDF.
Being built; not open source yet
Event Management
in progressA messaging bot for events.
Not running today
Research · 2
Captain Cook
researchAn agent pipeline whose work you can audit: plan, gate, worker, ledger.
Research; our n8n workflows move here
Open source · MIT
Video from July 2026
Agent Farm
researchTeams of agents built from templates that work on a task together and hand back their results.
Part of the Captain Cook project; no runtime yet
How a space works — and why it is open
Vibemind is an open-source operating system for a fleet of AI agents. Two ideas carry the whole system: spaces bound what can happen, and agents do the work inside them. You type or message an intent; a deterministic router called the Brain decides who handles it, an agent runtime executes it, and two knowledge graphs remember what was decided and what your material actually is.
It is a harness you can read. Routing, execution, memory and transport are separate, inspectable subsystems rather than one long prompt, and the capability registry tells you what the system can do before a model is ever asked.
It runs on your own hardware, the runtime sends no telemetry, and model calls go to whichever provider you configure. The public repository, vibemind-space/vibemind-os, is MIT licensed. Where a subsystem is unfinished, removed or unverified, the documentation says so with a status badge, and the claims register records every public claim we have had to correct.
18 people. One mission.
Three co-founders, three leads and four teams building Vibemind — most of us in Germany, our CEO in Lebanon.
Felix Baumann
Co-founder, CTO · Munich
7+ years building automation and AI systems at BMW, Micron, and IAV. Architects the multi-agent engine, the OpenClaw desktop driver, and the MCP integration layer powering every Vibemind space.
Stéphane Boghossian
Co-founder, CEO · Lebanon
10+ years in GTM and product marketing (UAE government, Lovable, Brainboard). Leads vision, narrative, and design-partner strategy.
Sophie Simic
Co-founder, CMO · Munich
Owns marketing and the quality bar for what ships. The voice of Vibemind in front of investors and partners.
Leads
Surya
COO · Munich
Runs operations. Bridges the open-source contributor community and the core team, and keeps a distributed team coherent.
Bora
Head of Sales · Munich
Leads sales — outbound, partnerships, and getting Vibemind in front of the right early users.
Samu
Head of Machine Learning · Munich
Machine learning specialist. Leads the machine learning work at Vibemind.
Vibe Coding
Sabrina
Vibe Coding · Robotics · Germany
Master's in robotics. Bridges Vibemind's agent layer with real-world embodied automation.
Alex
Vibe Coding · Technical Team Lead · Germany
Experienced in leading teams. Leads on the technical side and advises across the company.
Muttu
Vibe Coding · Germany
Homelab & Infrastructure
Andreas
Homelab & Infrastructure · Germany
Christian
Homelab & Infrastructure · Germany
Runs servers, runtime, and the operational backbone that keeps multi-agent workloads humming.
Moritz
Homelab & Infrastructure · Germany
Sales, Business & Finance
Steffen
Sales · Vibe Coding · Germany
Helps with sales, alongside his role at another company.
Nevin
Sales & Business Administration · Vibe Coding · Germany
Joy
Sales & Business Administration · Vibe Coding · Germany
Arjun
Business · Germany
Nadine
Finance · Germany
Ethics
Lisa
Ethical Guidelines · Germany
Working on ethical guidelines and responsible-AI principles for Vibemind. A commitment we are building toward, not a finished framework.
Partner
Hazem
Partner · with his own product · Germany
3
Co-founders
4
Teams
18
People
Where it actually stands
A running system, mapped honestly: runtime checks as of 27 October 2026, space labels as of 11 November 2026. Several spaces run on our own server; the central Brain service is being brought back into the path. The work now is coverage, cleanup, and shipping the first open release.
Running today
- Five spaces on our own server: Sales Claw, Marketing Agent, Business Data Layer (Rowboat), Video (Laura), MiroFish
- OpenFang agent runtime with 43 agents
- Agent task chain: order → agent → check → delivery, verified end to end
- Regex routing with ground-truth checks against the live database, verified
- Telegram in and out through OpenFang, verified end to end
- The 3D workspace in Electron and the security proof suite
In progress
- Bringing the Brain service back into the request path
- Video and marketing: moving generation from a workstation to the server
- Wiring the Brain directly into voice, with push-to-talk
- Brain coverage for every live space, in EN and DE
- Desktop automation, coding engine, research, scheduling and SWE spaces
- Codebase and licence cleanup ahead of the open-source release
Next
- Local models and home-server setup
- Agents on demand, spun up per task
- Autonomous deployment
- Hosted cloud, so you don’t need the GPU
- Community spaces and multi-language support
We'd rather tell you what's early than have you find out. If a claim on this page matters to your decision, the code is open — go read it.
Rather learn how to build this yourself?
AI consulting and coursesQuestions people ask
- Which spaces can I use today?
- Five spaces run on our own server today: Sales Claw, Marketing Agent, Business Data Layer (Rowboat), Video (Laura), MiroFish. Eleven more are in progress, and Captain Cook and Agent Farm are research. Each space works on its own or as part of Vibemind. The homepage labels every space and says which ones are open source; you can request any of them by e-mail. Read more →
- Does Vibemind offer AI courses or consulting?
- Yes. Courses in three levels (beginner, advanced and pro) on Claude Code, MCP, agent workflows with n8n and AutoGen, safe AI and evaluating agent frameworks, plus 1:1 sparring on your own project. The first 30 minutes are free and prices are on request. Courses run in the Vibemind Academy, in English and German. Read more →
- What is Vibemind?
- Vibemind is an open-source operating system for a fleet of AI agents. Spaces bound what can happen and agents do the work inside them. An intent you type or message is routed by a deterministic component called the Brain, carried out by agents in the OpenFang runtime, checked against ground truth, and remembered by two memories: one for what the system decided, one for what your material is. Read more →
- Is Vibemind open source?
- The umbrella repository, vibemind-space/vibemind-os, is public on GitHub under the MIT licence. The wider stack is assembled from components under several licences: the OpenFang runtime is Apache-2.0 or MIT, several libraries are MIT, some components carry no licence file yet, a few are AGPL-3.0, and one brain component and several AI models are licensed for non-commercial use only. One video component is private. The licensing page lists every component. Read more →
- Can I self-host Vibemind, and what do I need?
- Self-hosting is the point of the project. It expects a container runtime, Linux as the deployment platform, and a GPU if you want local models and embeddings; otherwise it uses hosted model providers. A vector store and a document store ship with the stack. It is not a packaged one-command install today: a small team runs it on their own machines, and you should expect to read configuration. Read more →
- Does Vibemind send telemetry or keep my data?
- There is no telemetry in the runtime; that is a design decision. The knowledge vault is plain files in a git repository on your machine, generated code goes to your own git, and model calls go wherever you configure them. Today most agents use a hosted Claude model; a local runner is possible if nothing should leave the machine. This website, unlike the runtime, does count visits with an analytics service; the privacy page says what it records. Read more →
- What is a space?
- A space is a bounded working context with its own agents, tools and vocabulary. You do not switch applications; you switch spaces, and the set of things the system can do changes with you. Scoping tools to a space bounds the blast radius of a bad decision and makes routing tractable: disambiguating against twelve candidate tools works, against five hundred it does not. Read more →
- How does Vibemind decide which agent handles a request?
- Deterministically first. The Brain matches the request against a registry of 119 capabilities using regular expressions; semantic matching with a model is meant as the fallback and is not working yet. A registry hit is free, instant and explainable, which is why the interesting behaviour of the system is readable as configuration rather than emergent from a prompt. Read more →
- Which parts of Vibemind work today?
- Each documentation page carries a status badge: running, partial, untested, removed or planned. The OpenFang runtime and the MCP tool layer are marked running, and a Telegram message to the bot was answered end to end on 27 October 2026. The Brain is partial: its routing and its agent task chain were verified end to end on 27 October 2026, but the Brain service itself is scaled down for now. The memories, the video space and most other spaces are partial; voice, scheduling and the coding engine are untested. Five spaces run on our own server today, two of them still with generation on a workstation. The status page lists every subsystem with what can actually be verified about it, and the claims register records every public claim the site has had to correct. Read more →