How it works
Own your workflow.
Own your AI bill.
Aurora is open source under MIT. Use the hosted version, or run it yourself — and bring your own OpenRouter key so AI costs stay yours, not a vendor's revenue line.
The 4-step setup
From signup to flow, fast
Same path whether you're hosted or self-hosted.
Pick how you run it
Deploy it anywhere: git-clone and ship to your own infrastructure. Identical codebase — switch later without a migration.
MIT licensed. Cloudflare Workers, Vercel, Fly, your own box — anywhere TanStack Start runs.
Create a workspace
Workspaces hold projects, custom fields, workflow templates and people. Roles and row-level security wire in on first commit.
Invite your team, set roles, pick a workspace preset (agency, product, consulting) and you're seeded with sensible defaults.
Drop in your OpenRouter key
Settings → AI → paste your key. Meeting analysis, task agents, Magic Plan, automations turn on instantly. Skip it and the rest of the app still works.
Pick any model OpenRouter supports — GPT-5, Claude, Gemini, Llama. Switch per-feature.
Work the way you think
The same tasks render as Table, Board, Canvas, Calendar or Timeline. Switch perspectives instantly — your data is the source of truth, not the view.
Saved views per person, per project. No more 'export to spreadsheet to actually look at it.'
Under the hood
A small, modern stack you can actually read
No microservice maze, no proprietary runtime. Open source primitives all the way down.
React 19 · TanStack Router · Tailwind
TanStack Start · Cloudflare Workers
RLS · realtime · your model
TanStack Start v1
React 19, file-based routes, typed server functions. SSR + edge-ready.
Supabase Postgres
Row-level security from day one. Migrations versioned in the repo.
OpenRouter
One key, 100+ models. You pay the provider rate, no markup.
Cloudflare Workers
Edge runtime by default. Sub-100ms server functions globally.
Your data, your rules
Self-host means the database lives in your account. No vendor lock-in.
Open source, MIT
Fork it, extend it, ship your own SaaS on top. PRs welcome.
Self-host
Five commands. Your own Aurora.
Runs on Cloudflare, Vercel, Fly or any Node host. Detailed guide in the README.
$ git clone https://github.com/zenifold/aurora-os.git aurora$ cd aurora && bun install$ cp .env.example .env # add Supabase URL + anon key$ bunx supabase db push$ bun run build && bun run deployBring your own key
The cost math nobody else shows you
Most tools mark up AI 3–10× and charge per seat on top. Aurora connects to OpenRouter — you pick the model, you pay the provider rate.
| Tool | Price | 10-person team / yr | Note |
|---|---|---|---|
| Notion AI | $10 / seat / mo | $1,200 / yr | Per-seat, model locked |
| ClickUp Brain | $7 / seat / mo | $840 / yr | Per-seat, model locked |
| Asana AI | Enterprise add-on | $2k+ / yr | Bundled, opaque pricing |
| Aurora + your key | Actual model cost | ~$60–$240 / yr* | Shared, any model |
*Estimated for a 10-person team running daily meetings, Magic Plan and agents on Gemini Flash / GPT-5-mini. Heavy GPT-5 use is more; the point is you can see and pick.
Security & data
Yours stays yours
Row-level security
Every table ships with RLS policies. Nobody — including the hosted operators — sees what they shouldn't.
BYO key, BYO data
Self-host means your Postgres, your secrets, your audit log. Hosted runs on the same code with a clear privacy line.
No AI markup
OpenRouter bills you direct at provider rates. We don't see your usage, we don't see your bill.
“We replaced Notion AI, Linear, and a Zapier subscription with one self-hosted Aurora. Our AI bill went from $480/mo to about $14.”
Questions
Things people ask
Do I need an OpenRouter key to use Aurora?+
No. Tasks, projects, CRM, meetings recording, notes, the whole workspace works without it. The key unlocks AI features: summaries, agents, Magic Plan, draft replies, transcript analysis.
What's the difference between hosted and self-hosted?+
Same code, same features. Hosted is one click — we run the database, edge functions and updates. Self-hosted is your infra, your database, your control. You can move between them without losing data.
How does the AI cost math actually work?+
OpenRouter bills you per-token at the provider's published rate. A typical 10-person team running daily meetings, agents and Magic Plan lands around $5–20/month total — not per seat. Heavy GPT-5 use is more; Gemini Flash is pennies.
Is my data sent to OpenRouter?+
Only the data needed for the specific AI call (e.g. a meeting transcript when you click Summarize). No background training, no telemetry to us. Self-hosted means even less surface area.
Can I bring my own model provider?+
OpenRouter is the default because it routes to everyone in one key. Direct OpenAI/Anthropic/Google keys are on the roadmap — and trivial to wire if you're self-hosting.
What happens to my data if I stop using Aurora?+
Export everything as CSV/JSON from Settings. Self-hosters already have the raw Postgres database. There's no proprietary format and no vendor lock-in.
Try it now, host it later
Start hosted in 30 seconds. Move to your own infra whenever you're ready — same data, same code.