Personal AI Ecosystem

I wanted a personal AI. So I started building the stack.

Not another chatbot. Not another collection of subscriptions. I wanted an AI that could know me, remember, work beside me, attend meetings, help build software, use compute I control, and eventually share both work and life with me.

0:00 / 10:12

Why this exists

The problem is not a lack of AI. It is fragmentation.

Chat, meeting notes, coding, planning, calendars, knowledge, voice, avatars, and model access often live in separate products. Each company owns another slice of context, charges another subscription, and stores another piece of identity. I want something different.

Instead of isolated assistantsOne AI identity
Instead of scattered contextOne memory architecture
Instead of a single providerCompute I can route and control
Instead of one rigid interfaceMultiple places to meet the same AI

“The pieces already existed. The difficult part was making them behave like one system.”

System architecture

Five projects. One personal stack.

HomePilot holds the center: identity, memory, knowledge, policy, and orchestration. The other projects give that intelligence a workplace, a presence, compute, and a way to help build itself—with me in control.

Different surfaces

One intelligence. Different places to meet it.

I do not want a DayPilot AI, an Avatar AI, a Meeting AI, and a Coding AI that behave like strangers. I want the same underlying identity and continuity wherever the conversation happens.

  • At workDayPilot organizes the professional workspace.
  • In meetingsMeetingSense brings context into HomePilot.
  • While codingGitPilot turns intent into reviewed changes.
  • At homeThe Avatar gives the same persona presence.
  • On any machineOllaBridge decides where inference runs.

A day with the system · vision

The interface changes. The relationship and memory do not.

Morning briefing

HomePilot brings the priorities and context. DayPilot presents the workday.

Deep work

GitPilot helps implement a feature, using a model routed through OllaBridge.

Meeting

MeetingSense is designed to listen, transcribe, understand shared material, and track actions.

Follow-up

HomePilot could surface three decisions, two actions, and one unresolved question—then ask permission before creating anything.

Presentation

Future presenter and coach capabilities can prepare context and talking points.

Work ends

DayPilot goes quiet. Professional tools recede.

Personal time

The same persona appears through the 3D Avatar for voice, cooking, language practice, exercise, a movie, VR, or conversation.

Long-term direction

From chatbot to personal secretary.

The goal is not a model that only waits for prompts. It is an AI that can know what is happening, remember what matters, prepare me, work with me, and follow up—while asking before any consequential action.

VISION · NOT A RELEASE CLAIM
“Good morning. You have three meetings today. The pricing proposal from last week is still open. I prepared the previous decisions and latest document. Would you like a five-minute briefing?”

Afterward: “Three decisions were made. You own two actions. Shall I create the tasks and draft the follow-up?”

GitPilot × OllaBridge

The stack helps build the stack.

I can use my own AI infrastructure to help build my own AI infrastructure. This is not autonomous recursive self-improvement. It is an engineering loop with an explicit human checkpoint.

IdeaExplorerPlannerCoderTestsReviewerHuman approvalRepository
Compute beneath the loop: OllaBridge → local GPU → workstation → optional provider. Commercial models remain useful, but they are no longer the only path.

Local-first, not cloud-absolutist

Local where practical. Cloud where useful. Choice always.

CONTROL

I decide where models and data run.

PORTABILITY

Applications depend on open interfaces instead of one vendor.

OPTIONALITY

Commercial providers are used when they add value, not because the system has no alternative.

I do not want zero cloud.
I want zero forced cloud.

Local-first and open-source by default makes paid AI optional rather than mandatory. Hardware, electricity, optional APIs, and cloud compute can still cost money.

What this solves

Replace seams with architecture.

Fragmented AI identity→ HomePilot persona
Meeting knowledge disappears→ MeetingSense + retrieval
Work context is scattered→ DayPilot
Every app needs a model endpoint→ OllaBridge
AI coding is tied to one provider→ GitPilot
AI has no presence→ 3D Avatar
Recurring subscriptions multiply→ Local/open alternatives where practical
Private context is split across vendors→ Self-hosting + explicit permissions

The projects

Independent today. Designed to converge.

These are active repositories at different stages. The cards separate the role each project already plays from the larger integration direction.

BRAIN · CORE

HomePilot

HomePilot is who my AI is. The personal AI backend for personas, identity, memory, knowledge, multimodal capabilities, tools, communication, policy, and model access.

  • Persistent personas
  • Long-term memory
  • Knowledge / RAG
  • MeetingSense
  • Agent orchestration
Current foundation · broader secretary experience is vision

View HomePilot on GitHub ↗

WORK

DayPilot

DayPilot is how my AI works with me. A professional cockpit where HomePilot personas can operate with context and approval.

  • Calendar & tasks
  • Projects & documents
  • Planning & email
  • RAG & agents
  • Approval center

View DayPilot on GitHub ↗

PRESENCE

3D Avatar Chatbot

The Avatar is how my AI shares space with me. Embodiment, voice, expression, and a future social and leisure surface for the same persona.

  • 3D avatar & voice
  • Expressions
  • WebXR / VR / AR
  • Companion mode
  • Coaching & activities
Active development · shared HomePilot identity is direction

View development branch ↗

COMPUTE

OllaBridge

OllaBridge is where my AI thinks. One OpenAI-compatible control plane between applications and the place inference actually runs.

  • Local Ollama / GPU
  • Remote workstation
  • Optional cloud GPUs
  • BYOK providers
  • Model routing

View OllaBridge on GitHub ↗

BUILDER

GitPilot

GitPilot helps me build the ecosystem itself. Specialized agents explore, plan, generate, test, and review—with approval modes and GitHub integration.

  • Repository exploration
  • Plan & code generation
  • Tests & review
  • Pull requests
  • Local model support

View GitPilot on GitHub ↗

The manifesto

OWN THE AI.OWN THE MEMORY.OWN THE COMPUTE.OWN THE TOOLS.

  • My AI should remember me because I choose it to.
  • My data should not require a SaaS account to exist.
  • My applications should survive a change of model provider.
  • My GPU should be useful when I already own one.
  • An AI should ask before taking consequential actions.
  • Work AI and personal AI should not require separate identities.
  • Open source should make the system inspectable and adaptable.
  • Local-first does not mean isolated.
  • Automation should increase my control, not remove it.

Where this is going

Capability stages, not release promises.

NOW

Core applications

The projects exist independently and continue to evolve.

CONNECTING

Shared foundations

Personas, model routing, meeting intelligence, and tools converge.

WORKING TOGETHER

Daily secretary

DayPilot and HomePilot form a continuous work system.

PRESENT

Voice and presence

The Avatar gives the same intelligence an embodied surface.

AMBIENT

Available, not intrusive

The AI becomes reachable across devices while respecting boundaries.

Build in public

Built for myself. Open so others can inspect.

This is a personal engineering project and a long-term vision—not a pitch deck. The repositories are public so people can learn, fork, improve, or use only the pieces they need.

The reason

I am not trying to build another chatbot.
I am trying to build the AI I wanted to have.

One that knows my work. Remembers what matters. Runs on infrastructure I can control. Helps build the software around it. And eventually shares both productive and ordinary moments with me.

— Ruslan Magana Vsevolodovna