Illustrative call · German
AI that works
inside your business.
An AI automation business from Hamburg. We make AI work inside your business in three ways.
The company film · 1 min 48 s
This is AuraMind.
An AI automation business from Hamburg.
One week at a car workshop: AuraMind Voice answers the phone, your own AI answers the team, and automations carry the routine.
Explore a sample callCar workshop · Salon · Property+
Illustrative scenarios
“Could I book a brake inspection?”
The agent understands the request, offers a time and waits for agreement before confirming.
Illustrative example. No real appointment is created.
Three ways.
One business.
1 AuraMind Voice answers your phone. 2 Your own AI runs on your own server. 3 Automations work for your team. Software is built around it.
Explore the three partsGive your AI tools
a job worth doing.
Choose a tool. Watch an example. See what we set up for your team.
OpenClaw
Open-source AI assistant that runs on your own devices.
Turn a brief into a content draft.
An agent can gather approved information and prepare a draft. Your team decides what gets published.
Enlarge reference imageIllustrative workflow · sample data
Give it a brief
Prepare three posts for our new service. Use only the approved product notes.
Set the boundaries
Product notes in. Brand guidance attached. Publishing permission stays with a person.
Prepare the draft
The agent assembles a content draft and marks information that still needs checking.
Review before publishing
Three draft posts, source notes and open questions arrive in your review queue. Nothing is published by this example.
- Runs on
- Self-hosted: your computer, server or private cloud.
- Connects to
- WhatsApp, Slack, Telegram and the tools you allow.
- A person approves
- Before anything is published or sent.
- Limit
- Tools act on the host unless it is sandboxed.
Hermes Agent
Open-source agent for research tasks, run on your own machine.
Bring back research you can check.
Give a recurring research task a defined scope. Get a brief with sources, uncertainty and a next step.
Enlarge reference imageIllustrative workflow · sample data
Ask a focused question
Summarise changes in the suppliers we already follow.
Choose the sources
Use the agreed source list. Keep dates and original links with every finding.
Assemble the brief
Group relevant changes and flag missing or conflicting information.
Make the decision
Your team checks the evidence and decides whether any follow-up is needed.
- Runs on
- Self-hosted on Linux, macOS or Windows.
- Connects to
- Chat, email, web search, MCP tools and your chosen model.
- A person approves
- A person reviews the brief before anyone acts.
- Limit
- Approval covers risky commands, not every action.
Grok Bot / Grok Build
Hosted agent from SpaceXAI with its own cloud computer; Grok Build is its coding tool.
Hand over the follow-up. Keep the approval.
Grok Bot is a hosted teammate from SpaceXAI, formerly xAI. It gets its own cloud computer, signs into the tools you allow, finishes a multi-step job and checks back for approval. Grok Build is the coding agent in the terminal for building and repairing internal tools.
Enlarge reference imageIllustrative workflow · sample data
Give the Bot a job
Follow up on last week’s open quotes. Draft the replies, do not send them.
Allow the tools
The Bot signs into the shared inbox and the CRM on its own cloud computer. Sending stays switched off.
Work through the list
It reads each thread, checks the customer record and drafts a reply for you to check.
Approve before sending
Drafts and record updates wait for a person. Nothing is sent by this example.
- Runs on
- Grok Bot: the vendor’s cloud computer, not self-hosted. Grok Build: your computer, with model calls to the vendor’s cloud. Apache-2.0.
- Connects to
- The inboxes, CRMs and websites you let the Bot sign into; those business logins live on that cloud machine. Grok Build: your repository.
- A person approves
- The Bot checks back when it needs approval. A person reviews every Grok Build change before release.
- Limit
- Grok Bot is a beta for SuperGrok and Cursor subscribers. After Grok Build’s July 2026 upload incident, telemetry and upload settings are checked before a client repository is connected.
Meta Muse
Meta’s agent for small businesses on Facebook and Instagram.
Set a goal for the week. Approve the result.
A personal agent that works on a goal over time. Muse for Small Business connects Facebook and Instagram business accounts, Meta ad accounts and tools such as Shopify, Intuit QuickBooks, Stripe, Slack, Notion, Canva, Asana and Zoom. In Meta’s own words: “nothing publishes, sends, or spends without your approval.”
Enlarge reference imageIllustrative workflow · sample data
Set the goal
Review last week’s posts and ads for our shop. Draft next week’s campaign.
Connect the accounts
Facebook and Instagram business accounts, the Meta ad account and Shopify, each with only the access the job needs.
Work over time
Muse works on the goal on its own cloud machine and prepares a campaign draft with the numbers behind it.
Owner approves
Nothing publishes, sends or spends until the owner approves. This example publishes nothing.
- Runs on
- Meta’s cloud, a virtual machine per user. Not self-hosted.
- Connects to
- Facebook and Instagram business accounts, Meta ad accounts, Shopify, QuickBooks, Stripe, Slack, Notion, Canva, Asana and Zoom.
- A person approves
- Meta’s wording: nothing publishes, sends or spends without your approval.
- Limit
- US and Canada at this check, not Germany. Queries are used for model training unless switched off.
Claude Code
Anthropic’s coding agent for building and changing software.
Build a tool around a real task.
Turn a repeated spreadsheet task into a reviewed software change, with an agreed plan and checks.
Enlarge reference imageIllustrative workflow · sample data
Describe the task
Combine our weekly order exports into one review screen.
Plan the change
Inspect the project and define accepted input, output and error handling.
Build and test
Prepare code changes and run the agreed checks against sample files.
Review the result
Inspect the change, try the workflow and approve the release separately.
- Runs on
- Your computer or the browser. Model calls go to Anthropic.
- Connects to
- Your repository, tests, GitHub and MCP tools.
- A person approves
- Set to ask before edits and commands. A person reviews the change.
- Limit
- Undo covers file edits, not outside systems. Paid plan needed.
Codex / Automations
OpenAI’s coding agent, plus scheduled tasks that run on their own.
Find the fault. Check the fix.
Investigate a broken internal workflow, make a scoped change and show the test results. Automations then run tasks on a schedule or from supported app events: a daily brief, a weekly summary, a cleanup. A person reviews each result.
Enlarge reference imageIllustrative workflow · sample data
Show the failure
Our import creates duplicate contacts when a file is retried.
Reproduce and fix
Inspect the import path with safe sample records, prepare a focused fix and rerun the checks, including the retry case.
Schedule the follow-up
An automation reruns the import check each morning and writes a short brief. A weekly summary lists what changed.
Inspect before release
Your team receives the proposed change, the test evidence and each scheduled brief. Deployment is a separate decision.
- Runs on
- Command line, IDE, desktop app or OpenAI’s cloud. Team Tasks run in the cloud through a service account.
- Connects to
- Your repository, scripts, CI and MCP tools. Automations start on a schedule or from supported app events.
- A person approves
- A person checks the diff, the test results and what each scheduled run produced.
- Limit
- Paid ChatGPT plan or API key. A local automation needs the app open. Network is off in its sandbox by default.
n8n
A workflow tool that connects your apps and moves data between them.
Move enquiries to the right next step.
Connect your form, records and response draft. Missing information goes to a person instead of disappearing.
Enlarge reference imageIllustrative workflow · sample data
Receive an enquiry
A customer asks for a quote through your existing form.
Check the details
Validate required fields and check whether this enquiry already exists.
Route the work
Create or update the record and prepare the next task for the right team.
Review the response
A response draft awaits approval. Incomplete requests stay visible in a handoff queue.
- Runs on
- Your server (for example Docker) or n8n’s cloud.
- Connects to
- Hundreds of business apps, webhooks, forms and AI models.
- A person approves
- The workflow can pause until a person approves.
- Limit
- Licensed for internal business use. Alerts only if configured.
Open Harness
AuraMind’s own workspace where several AI tools work on one task.
Keep the discussion and build together.
AuraMind’s working prototype keeps model discussions, an approved plan and build checks in one workspace.
Enlarge reference imageIllustrative workflow · sample data
Bring a question
Compare approaches before changing our project.
Compare perspectives
Selected models respond independently and review a shared proposal.
Approve and build
An agreed plan goes to an eligible coding connection. Verification runs separately.
Inspect the evidence
Review the recorded discussion, changes and check results before accepting the work.
- Runs on
- Your Mac, as a desktop prototype.
- Connects to
- Claude Code and Codex, with their own sign-ins.
- A person approves
- You approve the plan, then review the result.
- Limit
- Working prototype. Falls back only to the same model.
Examples show a possible setup. Accounts, permissions and approvals are set up for your business. View all tools and integrations
Built here.
Ready to explore.
Actual prototype screenWorking prototype
Open Harness
Let models compare ideas. Keep the plan, build and checks in one workspace.
See the workspaceSpeak. Capture. Review.
⌥PI
In development
Local Dictation
Explain it with your voice and screenshots. Review the message before it leaves your hands.
Explore the prototype2 · Your own AI
Run local AI models
on your own business infrastructure.
An open AI model runs on a server or Mac inside your company. We connect it to your documents and tools, and your team asks it questions like a chat assistant. Where it pays, we fine-tune it on your business. With cloud features off, nothing leaves your network.
Company assistantInternal app · uses your own model
Employee · Company assistantWhat is our warranty on brake pads?
Your local model · gemma4:31b24 months or 40,000 km, whichever comes first.
Source: Service manual, page 12
Website enquiry · n8n workflowCan you service our two vans on Saturday?
Your local model · gemma4:31bYes. Saturday has two free slots, 9:00 and 11:00. Draft reply ready, waiting for your OK.
Source: Opening hours and price list 2026
Supplier invoice · n8n workflowCheck invoice 4471 against the order.
Your local model · gemma4:31bAmount and VAT match the order. The delivery date differs by one day, so it goes to a person.
Source: Order 4471 and the invoice PDF
Company server · Ollama
$ ollama ps
NAME ID SIZE PROCESSOR CONTEXT UNTIL
gemma4:31b 3f1e9c27a0b4 21 GB 100% GPU 32768 Forever
$ ollama serve
Listening on 10.0.4.10:11434 (version 0.12)
[GIN] 2026/10/12 - 09:14:03 | 200 | 1.9s | 10.0.4.21 | POST "/v1/chat/completions"- Model
- gemma4:31b · 4-bit
- Hardware
- One workstation GPU with 32 GB
- Address for your tools
http://ai.internal:11434/v1- Cloud features
- Off
- Data leaving your network
- None
What is Ollama?
Ollama is free, open-source software that runs an open AI model on your own Mac, PC or server. Your apps talk to it the same way they talk to OpenAI, so nothing has to be rebuilt. With its cloud features off, questions and answers stay on your machine.
When many people use it at once, the same models run on vLLM or SGLang on a Linux server with NVIDIA GPUs. We pick the runtime for your load.
Picture: the official Ollama app, ollama.com. No partnership implied.
The open models we run for companies.
All of them download through Ollama with one command and run without an internet connection. We test two or three on your real task and keep the one that answers best.
- Gemma 4Google DeepMind · Apache 2.0Documents and answers with the source; a good first choice for German-speaking teams.Runs on 12B on a 16 GB machine. 31B on one GPU with 24 to 32 GB.
- Qwen 3.5 · 3.6 · 3.8Alibaba · Apache 2.0Agents that use tools, coding help and multi-step tasks.Runs on 9B on 16 GB. 27B on one workstation GPU. 122B needs 48 to 96 GB.
- gpt-ossOpenAI · Apache 2.0Tool use, answers in a fixed format and short reasoning steps.Runs on 20B within 16 GB. 120B on one 80 GB GPU.
- DeepSeek-R1DeepSeek · MITStep-by-step checks where the reasoning should be visible.Runs on Distilled 7B to 70B on workstations. 671B on a multi-GPU server.
- Llama 3.1Meta · Llama community licenceA widely used baseline for drafts and summaries.Runs on 8B on 16 GB. 70B needs 40 to 48 GB.
Sizes and memory are approximate and depend on quantisation and context length. Checked against the official model cards and the Ollama library on 11 October 2026.
Why it works for a whole company.
- One server, many teams
Service, sales and accounting use the same model through one internal address. For many people at once, it runs on vLLM or SGLang on a GPU server.
- Your data stays inside
Documents, questions and answers stay on your hardware. Who may use the model and what is logged is set up and written down.
- Fixed costs, no per-request fees
You pay for the hardware and its upkeep, not for every question. We compare that with cloud API costs for your volume before you decide.
- Your tools stay the same
n8n, OpenClaw, Hermes Agent and your own apps call the internal address instead of a cloud API. Only the address changes.
A normal week
with AuraMind.
Four everyday situations from the film. Illustrative examples: a person approves before anything is posted, sent or paid.
1 / 4 · Social media
One photo of today’s tart.
A short video and two posts, ready to approve.
The bakery takes one photo with the phone. Your own model writes the captions in your tone, and a workflow prepares a short video, an Instagram post and a Facebook post. Nothing is published until you approve it.
- One photo from the phone
- A short video and two posts are drafted
- You approve, then they are scheduled
Your own model · OpenClaw or n8n · your scheduling tool
- Short video · 9 sSquare cut with the photo and your logoReady to approve
- Instagram postCaption in your tone, two hashtagsReady to approve
- Facebook postSame photo, opening hours addedReady to approve
ApproveEdit
2 / 4 · Enquiries
An enquiry at ten at night.
A reply from your price list by the morning.
The enquiry lands in your inbox at 22:04. Overnight your own model reads it, checks your 2026 price list and drafts the reply. In the morning the draft is waiting for your OK. Nothing is sent without it.
- The enquiry arrives at 22:04
- A reply is drafted from your price list
- It waits for your OK, then it is sent
Your own model · n8n · your inbox and CRM
Can you manage our two buildings from March?
Good morning. Yes, we can take over both buildings from 1 March. The rate from our 2026 price list and the next steps are below.
Source: Price list 2026Waiting for your OK
SendEdit
3 / 4 · Invoices
A supplier invoice is read,
checked against the order and entered.
The invoice PDF arrives. The fields are read, the total is checked against the purchase order and the entry is prepared in your accounting. Unclear values are flagged for a person. Payment stays with your team.
- The invoice PDF arrives
- Fields are read and checked against the order
- Entered; your team releases the payment
Document parser · your own model · n8n · your accounting
- Supplier
- Brake parts supplier
- Invoice no.
- 2026-0917
- Date
- 3 October 2026
- Net
- EUR 1,240.00
- VAT 19 %
- EUR 235.60
- Total
- EUR 1,475.60
Matches order #4471EnteredPayment: your team
4 / 4 · Team answers
A new colleague asks about the warranty.
The answer comes from your service manual.
The question goes into the team chat. Your own model answers from the service manual and names the page, so the colleague can check it. The knowledge base is shared, so every answer stays available to the whole team.
- A question in the team chat
- The answer comes from the service manual, with the page
- One shared knowledge base for everyone
Your own model · shared knowledge base · your team chat
What’s our warranty on brake pads?
24 months or 40,000 km.
Service manual, page 12
Technology chosen for the taskExplore the stack
Work you can
look into.
See all projects
Client work · Voice
Voice reception for garages
ProblemA garage wanted calls answered in French, with the vehicle, contact details and preferred appointment time captured.
ResultA configured reception assistant with instructions for appointment-request intake.
Client work · Local AI
Local agents on a Mac mini
ProblemThe client wanted agents running on their own Mac mini and needed to know what 16 GB of memory could handle.
ResultAn evaluated setup, installation script and practical operating guide.
Own build · Research
Research with traceable sources
ProblemResearch drafts needed sources a person could check before anything was saved.
ResultSource-linked research and German drafts saved for review.
Tell us the task.
We’ll map the next step.
We plan it with you, build it and keep it running. A missed call, a repeated task or a model you want on your own server: start with a short message.
Do I need to understand AI?
No. Tell us what takes time. We explain the options and help you choose a useful first workflow.
Can you connect my existing tools?
We assess your phone, calendar, CRM and other systems first. Your setup depends on the integrations and access they provide.
What does a setup cost?
We scope the work with you and explain setup and ongoing service costs before you commit.