Local-first agentic AI OS

Your own Jarvis,
running on your
own machines.

GauravOS is a command center for a fleet of AI workers — orchestrated by NOVA, planning, coding, testing and shipping across the machines on your own network. No cloud. Your data never leaves the LAN.

Private beta · self-hostable · view on GitHub
NOVA
THINKING
MACHINES
0
AI WORKERS
0
TASKS / DAY
0
RUNS ON
Your LAN
The command center

One screen for your entire AI operation

Live telemetry from every machine and worker, the second-brain graph, agent topology, and an evidence log of everything NOVA did — while you slept.

gauravos · localhost:3100NOVA CORE · ONLINE
Obsidian Second BrainSYNCED 12s
12,480 notes 41,022 links 318 indexed
cluster: project-alpha
Worker Fleet4 MACHINES · 6 BUSY
Orchestrator PM
forge-01 · routing task TASK-2041
Engineering
forge-01 · orders API slice
QA Browser
atlas-02 · Playwright e2e 9/14
Compression
vault-03 · Headroom ~61% ctx
> running vitest ✓ 128 passed
> POST /api/orders 201 84ms
> persist audit_event
● NOVA CORE · ONLINE
uptime 41d 06:12
Compressing vault context · Headroom pass
Event Stream · Evidence Log
21:52:01CompressionHeadroom −61% before model call
21:51:58QAPlaywright e2e 9/14 passed
21:51:58Engineeringcommit a3f² · orders API
21:51:52Securitydependency scan clean
21:51:44Approvalsspend $12.40 auto-approved
01 — NOVA Core

The orchestrator that never sleeps

NOVA is the brain of the OS — a supervisor model that takes a goal in plain language, breaks it into a plan, and dispatches the right specialist worker for each step. Ask it to ship a feature and it plans, writes code, runs tests, reviews the diff, and logs every decision.

Talk or type
"Ask NOVA or issue a command" — natural language into an executable plan.
Plans, then delegates
Routes each step to Engineering, QA, Research, Security or Deploy workers.
Context compression
A dedicated worker keeps the context window lean so long jobs never stall.
NOVAqwen2.5-coder:7b · local
Ship the orders API and get it under test.
Planned 4 steps. Dispatching:
✓ Engineering — scaffold /api/orders
✓ Engineering — payments client
▸ QA — Playwright e2e (9/14)…
· Security — dependency audit scan
NOVA is executing…
Ask NOVA or issue a command…⌘K
Agent Topology · Supervisor
PM Orchestrator
ResearchEngQASecDeploy
Ryzen 9 64GB · RTX 4090 ·
Ryzen 9 · 64GB · RTX 4090 · 72%
M2 Ultra 128GB ·
M2 Ultra · 128GB · 41%
Xeon 256GB ·
Xeon · 256GB · 88%
Mac mini 32GB ·
Mac mini · 32GB · idle
02 — Worker Fleet

Every machine on your network, working as one

Register any Windows or Mac box as a worker with one command. GauravOS pools their CPUs, RAM and GPUs into a single fleet — then routes each task to the machine and model tier that fits, from a 7B local model to a frontier call when it's worth it.

11 roles
Orchestrator, Eng, Research, QA, Design, Security, Deploy…
3 tiers
Local · mid · frontier — cost-routed per task.
register-machine.ps1 -Server http://192.168.1.226:3100
03 — Second Brain

A memory that grows with everything you do

Point GauravOS at your Obsidian vault and it becomes shared, searchable memory for every worker. Notes, links and documents are embedded into a pgvector index — so NOVA answers with your context, cites the note it used, and gets sharper the more you write.

0
notes
0
links
0
indexed clusters
Pricing — cost modelProduct spec v3Security · access control
04 — Agent Lifecycle

Discover → Assign → Execute → Verify → Learn

Every task moves through the same governed loop. Nothing ships unverified, and every run feeds the second brain so the next one is faster.

Discover
Parse the goal, pull vault context
Assign
Route to worker + model tier
Execute
Run tools, write code, act
Verify
Tests, scans, approvals gate
Learn
Write results back to memory
05 — Workflows

Automations that run themselves

Schedule recurring jobs in n8n, wire them to NOVA, and watch success rates in real time. Approvals only surface when something's risky or costs money.

Lead enrichment
Every 15 min · enrich → CRM
98%
avg 412ms · last run 2m ago
Nightly deploy
02:00 UTC · build → e2e → ship
94%
avg 6.2m · last run 7h ago
Vault reindex
Hourly · embed new notes
100%
avg 1.1m · last run paused
06 — Local-first architecture

Runs entirely on your hardware

One docker compose up brings the whole stack online — Postgres + pgvector for memory, Ollama for local models, Open WebUI and n8n for chat and automation. No API keys required to start. Bind to your LAN and reach it from any machine in the house.

Encrypted at restRole-based accessFull audit logNo cloud egress
Postgres + pgvector
Persisted state & embeddings · :5438
Ollama
Local model runtime · :11434
Open WebUI
Chat surface · :8080
n8n
Workflow automation · :5678

Get your own Jarvis.

GauravOS is in private beta. Join the waitlist for early access, or book a live walkthrough of the command center.

Request a demoStar on GitHub