Custom Personal AI Assistant Setup
Privacy-first AI assistant built around how you work. Granular file access, swappable local/cloud models, internet toggle on demand, and repeatable custom workflows.
Why this isn't ChatGPT or Claude
ChatGPT and Claude are great - they're also generic, cloud-only, and forget who you are between sessions. This assistant runs on your hardware with your memory, sees only files you grant, and is configured around your specific workflows. Cloud models stay available as an opt-in fallback for heavier tasks - you decide per prompt whether a request can leave your machine.
You own the model
Runs locally - no rate limits, no provider lock-in, no telemetry.
You own the memory
Persistent files the assistant reads + updates over time. Yours forever.
You own the permissions
Granular file access + internet toggle. Cloud is opt-in per task.
What's Included
One-time fee. Everything below ships with your build.
Custom Personal AI Assistant
- Local LLM installation (Ollama, LM Studio, or your platform)
- Cloud model fallback setup (off by default, toggle per task)
- Internet-access toggle (no restart needed)
- Granular file-access rules
- Persistent memory + identity files
- Todo + notes + personal-knowledge integration
- Up to 3 custom workflows configured live
- Discord + Telegram messaging integration
- Google Calendar (read, add, edit, reschedule)
- Obsidian vault indexing + Q&A
- Home-server monitoring + permissioned remediation
- 90-minute live walkthrough + training
- 30 days of post-setup tuning
Four Pillars
Privacy, memory, integrations, and workflows - what makes this assistant yours.
Privacy Controls
Local by default. Cloud only when you say so.
- Runs locally on your hardware - no telemetry
- Internet access toggle - flip on or off without restart
- Granular file access - assistant sees only what you grant
- Hybrid local/cloud routing - opt-in cloud per task
Memory & Identity
It remembers what you want it to.
- Persistent memory files (the assistant's "soul")
- Identity file - voice, role, behavioral defaults
- Todo lists, notes, and personal knowledge integration
- Reads + updates memory across sessions
Integrations
Plugs into your existing stack.
- Discord + Telegram for chat-based briefings
- Google Calendar - read, add, edit, reschedule
- Obsidian vault indexing + Q&A
- Home-server monitoring + permissioned auto-remediation
Workflows
Repeatable automations on demand.
- Up to 3 custom workflows configured during setup
- Permission-gated actions - assistant asks before acting
- Pre-approve recurring actions for autonomous execution
- Add more workflows during the 30-day tuning window
Example Workflows
First 3 workflows are configured during setup. More can be added in the 30-day tuning window.
Morning Brief
Optimal todo list, weather, and commute time pushed to your Discord or Telegram every morning at the time you set.
Calendar by Text
Add, reschedule, or cancel calendar events with a single chat message - natural language, no UI to navigate.
Obsidian Q&A
Ask questions about your vault and get cited answers with note links. Summarize meeting notes or search across years of writing.
Server Monitoring
Pings you when a site goes down, dashboard crashes, or a Minecraft server maxes RAM. Asks permission, then auto-remediates.
Scheduled Check-ins
Health, progress, or habit check-ins at the cadence you set. Logs responses to memory for trend tracking.
How the Privacy Lanes Work
Two lanes, one master switch. Flip on the fly per session, per task, or per workflow.
Local Lane
Model runs on your machine. No data leaves your hardware.
- • Ollama / LM Studio / your platform
- • Sees only granted files
- • No internet required to function
Cloud Lane
Routes to OpenAI / Anthropic / your choice when enabled per task.
- • Use for heavier reasoning
- • Off by default
- • Toggle per prompt or workflow
Internet Toggle
Master switch over both lanes. When off, assistant is airgapped.
- • One toggle, no restart
- • Only allowed integrations stay live
- • Use for sensitive sessions
Hardware Requirements
Lighter hardware works too - cloud fallback covers what your local lane can't run.
| Component | Recommended | Minimum |
|---|---|---|
GPU | NVIDIA, 16GB+ VRAM | NVIDIA, 8GB VRAM (lighter models) |
CPU | 8+ cores (concurrent workflows) | 4+ cores |
RAM | 32GB+ | 16GB |
Storage | 100GB+ free (models, memory, logs) | 50GB free |
Your assistant can reach the tools you already use. Every integration is opt-in and permission-gated - it only does what you allow, and you can switch any of them off.
| Integration | How we connect it | What it enables |
|---|---|---|
| IMAP/SMTP or Gmail API | Triage, draft, send, and summarize mail; daily briefings | |
| Discord | Bot token on your server | Chat with your assistant and get push briefings anywhere |
| Telegram | Bot token via BotFather | The same assistant on your phone, no app to build |
| Slack | Slack app + bot token | A team-facing assistant in your workspace channels |
| Obsidian | Local vault indexing (RAG) | Ask across your notes; cited answers and summaries |
| Google Calendar | OAuth, read/write scopes | Add, move, and cancel events by chat message |
| Notion | Internal integration token | Read and update notes, tasks, and databases |
| Home Assistant | Long-lived access token | Status checks and permissioned device control |
| Webhooks / custom APIs | Inbound and outbound HTTP | Wire in anything with an API; trigger and receive events |
The Setup Process
Discovery to handoff in roughly 2-3 weeks depending on workflow complexity.
Discovery
Tell us your workflows, hardware, integrations, and what you want the assistant to remember. We confirm fit before booking.
Install + configure
Local LLM installed (Ollama, LM Studio, or your platform). Optional cloud fallback wired up. File permissions and internet toggle set.
Workflows + integrations
First 3 workflows configured live. Discord/Telegram/Calendar/Obsidian integrations connected. Memory + identity files seeded.
Walkthrough + 30-day tuning
90-min live walkthrough covering daily use, permission management, and adding workflows. 30 days of post-setup tuning included.
Runners like Ollama or LM Studio load and run the model. Assistant platforms like Hermes, Goose, and OpenClaw sit on top and add the memory, scheduling, and integrations below. Most setups use both - we pick the combination that fits your goals and hardware.
Tap any platform name for a quick description of where it came from and what it is good at.
| Capability | |||||
|---|---|---|---|---|---|
| Type | Omnichannel agent | Coding & automation agent | Local-first agent | Knowledge assistant | Model runner |
| License | MIT | Apache-2.0 | MIT | AGPL | Free, proprietary |
| Runs 100% on your hardware | |||||
| Chat apps (Telegram / Discord) | |||||
| Slack | |||||
| Calendar | |||||
| Files & notes (Obsidian vaults) | |||||
| Scheduled / unattended jobs | |||||
| Persistent memory & skills | |||||
| Built-in local model serving | |||||
| Best for | Living in your messaging apps | Dev workflows, automation, and MCP tooling | Hands-on automation that takes actions | Searching and acting on your notes | Simple, fast local model running |
These are examples, not an exhaustive list, and this space moves fast - the best harness and methods change with new releases. We re-assess and recommend the right fit during setup and your initial consultation. We are not affiliated with these projects.
The platform is the visible layer; these methods do the heavy lifting underneath. We pick and tune them to your needs.
- RAG over your files, backed by vector search, so answers cite your own documents
- MCP for integrations - the emerging standard for connecting tools to the model
- Sandboxed, permission-gated actions - the assistant only touches what you allow
- Local by default with opt-in cloud routing for heavier tasks
- Persistent memory so it keeps context across sessions
Already on a support plan?
Creator and Gamer Support Plan members: ask about bundled discounts on your Personal Assistant Setup. See support plans →
You Might Also Like
The foundation your Personal Assistant is built on - local LLM or full creator-tools package.
Learn MoreGo deeper on prompt engineering, workflow design, and getting the most out of your assistant.
Learn MoreNeed the hardware to run a powerful local model? Get a build tuned for AI workloads.
Learn MoreFrequently Asked Questions
Two things: ownership and control. Your assistant runs on YOUR hardware by default, with only the files and data you explicitly grant. Cloud models are an opt-in fallback for heavier tasks - you decide per prompt (or per workflow) whether a request can leave your machine. It's also configured around YOUR workflows and memory, not a generic chat interface.
Morning Discord/Telegram brief (todo + weather + commute), calendar management by text (add/reschedule/cancel events), scheduled check-ins, Obsidian vault search and summarization, home-server monitoring with permissioned auto-remediation. We set up the first 3 during setup and you can request more during the 30-day tuning window.
Your assistant has two lanes: a local lane (the model on your machine, no internet) and a cloud lane (routed to OpenAI / Anthropic / your choice when enabled). You can flip between them on the fly per session, per task, or per workflow. Internet access itself can also be toggled off entirely - when off, the assistant is airgapped except for integrations you've explicitly allowed.
For a good local experience: NVIDIA GPU with 16GB+ VRAM, 32GB RAM, 100GB free storage. 8GB VRAM works for lighter models. If your hardware is lighter than that, the cloud-fallback setup means the assistant still runs well - local lane just covers a smaller set of tasks.
Discord and Telegram (chat), Google Calendar (scheduling), Obsidian (notes), and home-server monitoring hooks (customizable per service). Additional integrations (Notion, Slack, Home Assistant, custom APIs) can be added during setup or in the 30-day tuning window.
30 days of post-setup tuning is included: refining workflows, adjusting permissions, adding integrations, troubleshooting. After that, additional tuning sessions can be booked, or you can roll into a Creator / Gamer Support Plan.
Yes - but only with your permission. Actions (sending messages, editing files, restarting services) are gated: the assistant asks first, and you can pre-approve specific recurring actions. You define what runs autonomously and what requires a check-in.
One-time fee of $250, including full configuration, 90-min walkthrough, and 30 days of tuning. No subscription. Additional tuning or new workflows can be booked later a la carte.
Ready to build your assistant?
Tell us about your workflows, hardware, and the integrations you care about. We'll confirm fit and timeline before kickoff.