Free personal AI · lives in your GitHub · yours forever

Personal AI.
Private by design.

Your own AI that runs on your machine, learns with you, and stores everything it learns in your private GitHub repo — not ours. Cerbral the AI is free; if you ever want more compute or storage, Cerbral Cloud has paid tiers on top.

or

By continuing you agree to the principles of Cerbral — we never own or sell your data. Only your email, GitHub handle, and brain-repo URL land in our database; everything else stays in your GitHub.

Already have an account? Sign in · Prefer the desktop app? Install Cerbral

Cerbral is free. Cerbral Cloud is optional.

You never pay for the AI itself — it runs on your Mac. Cerbral Cloud is a separate subscription if you want extra compute or storage on top; use as much or as little as you want.

Free

Your own Cerbral, on your own Mac.

$0

The AI itself is always free

Get Cerbral
  • Full Cerbral — same build, no "lite" edition
  • Runs entirely on your machine via local Ollama
  • Your brain repo is a private GitHub repo you own
  • Self-improves via the Hermes Agent loop
  • Desktop app + Cerbral Web (reads your local AI via a secure tunnel)
  • Works offline · phones home: never
Most popular

Cerbral Cloud · Pro

Cerbral on the web without your Mac on.

$20/mo

Or $192/year · save 20%

Start 7-day trial

No surprise bills · hard caps, not metered overage

Everything in Free, plus:

  • ~5M tokens of cloud inference a month — roughly 400 sessions, or 40 hours of Cerbral Code
  • $5 Boost credit for frontier models (Claude Sonnet, GPT) on demand
  • 10 GB Cerbral Cloud storage — brain-repo stays primary
  • Cerbral Web + mobile work even when your Mac is off
  • Scheduled + background agents hosted on Cerbral Cloud
  • Sync across devices · priority queue
  • Opt-in overage after cap: $3 per 1M tokens

Cerbral Cloud · Max

Heavy Cerbral Code usage, long-running agents.

$100/mo

Or $960/year · save 20%

Get Cerbral

No surprise bills · hard caps · cancel anytime

Everything in Pro, plus:

  • ~30M tokens a month (6× Pro) — roughly 2,400 sessions, or a full-time coding week every week
  • $25 Boost credit for frontier models — Claude Opus, GPT-Pro class
  • 100 GB Cerbral Cloud storage
  • Multi-hour background agents · longer run budget
  • Early access to new Cerbral models as Galactuz rolls them out
  • Lower queue times at peak load
  • Opt-in overage after cap: $2 per 1M tokens (bulk)

Cerbral Cloud · Ultra

Power users, teams of one, always-on agents.

$200/mo

Or $1,920/year · save 20%

Get Cerbral

No surprise bills · hard caps · cancel anytime

Everything in Max, plus:

  • ~60M tokens a month (12× Pro, 2× Max) — roughly 4,800 sessions, or 8-10 hours a day of heavy Cerbral Code
  • $50 Boost credit for frontier models · Opus + GPT-Pro unrestricted
  • 500 GB Cerbral Cloud storage
  • Dedicated inference slot · shortest queue at peak
  • Multi-day background agents · longest run budget
  • Earliest access to new Cerbral features before they hit Max
  • Opt-in overage after cap: $1.50 per 1M tokens (best bulk)

No surprise bills, ever. Every tier has a hard monthly cap by default — if you hit it, Cerbral Cloud stops charging compute until next month, full stop. Metered overage is strictly opt-in and capped at whatever monthly ceiling you set. In the app you'll see a progress bar and a plain-English estimate ("about 180 more sessions this month at your current pace") — not a token counter. You should never have to do mental math about tokens to use Cerbral.

Why the numbers hold up. Cerbral Cloud routes by default to open-weights models (Qwen3-class) hosted on Galactuz-managed GPUs, where token cost lands around $0.90 per million blended — so $20 Pro's 5M token budget costs us ~$4.50, leaving room for payment processing and margin. The Boost credit pool is what lets users reach Claude Sonnet or GPT-Pro on demand; those calls debit the pool at provider cost, so heavy frontier use doesn't torch the margin on the base sub. Storage is the cheap part: at Cloudflare R2 rates ($0.015/GB/mo), the 10 / 100 / 500 GB quotas cost us $0.15, $1.50, and $7.50 per user per month — negligible next to compute. Same framing Ollama uses for Turbo, with a Cursor-style pay-as-you-go on top. Your conversations and memories stay in your GitHub brain-repo; Galactuz never sees the contents.

For teams and companies

Same Cerbral product, shaped for multiple seats. Each member's brain repo stays private to them; the admin console, SSO, and one bill are what teams actually need from a cloud AI — not a shared token pool one person can drain.

Teams · Standard

Self-serve, for teams that want SSO + one bill.

$15 /seat · month · annual

$18/seat billed monthly · min 3 seats

Start a team

Pro, for every seat

  • Per-seat token budget — ~7M tokens/mo per seat (slightly above solo Pro)
  • $5 Boost credit per seat for frontier models — not pooled, not transferable
  • 15 GB Cerbral Cloud storage per seat
  • Shared Spaces for projects the team works on together
  • Admin console — invite/remove, usage per member, per-seat caps
  • SSO — Google Workspace, Microsoft Entra, domain capture
  • One consolidated bill · JIT provisioning · RBAC
Includes Cerbral Code

Teams · Premium

Heavy Cerbral Code usage per seat; unlocks long-running agents.

$60 /seat · month · annual

$72/seat billed monthly · min 3 seats

Start a team

Max, for every seat

  • Per-seat token budget — ~35M tokens/mo per seat (slightly above solo Max)
  • $25 Boost credit per seat for Opus + GPT-Pro class — per-seat, not pooled
  • 120 GB Cerbral Cloud storage per seat
  • Cerbral Code agentic mode unlocked on every seat
  • Multi-hour background agents · priority queue
  • Everything in Standard — admin, SSO, SCIM, RBAC
  • Opt-in overage per seat: $2 per 1M tokens

Enterprise

For companies that need a contract, audit trail, and BAA.

From $40 /seat · month

Min 25 seats · 12-month term · seat fee + metered usage

Contact sales

Everything in Teams Premium, plus enterprise controls

  • SCIM provisioning + deep SSO (Okta, Entra, custom IdPs)
  • Audit logs + Compliance API — programmatic access to every admin action, key event, usage record
  • Dedicated inference capacity — your own Galactuz GPU slice, no shared queue
  • HIPAA-ready BAA — sales-gated, not self-serve (on purpose)
  • Custom data retention + training opt-out by default
  • Data residency — US, EU, on-prem GPU hosting (1.1× for custom region)
  • 99.9% inference SLA with credits-back
  • Expanded context window · custom rate limits · named account team

Build on the Galactuz API

Pure usage-based inference for developers — the same routing stack Cerbral Cloud subscriptions use, with published per-token rates. Prompt caching, batch discounts, and usage tiers that unlock higher spend caps as you grow. $5 in starter credits for new orgs. Lives at platform.galactuz.com.

Default model (Qwen3-class)
$0.50 / $2.50
Per 1M input / output tokens · ~5× cheaper than Claude Sonnet. Same quality as Cerbral Cloud's default routing.
Frontier (Claude, GPT)
Pass-through + 15%
Anthropic + OpenAI published rates plus a thin routing spread. Opus, GPT-Pro, image+file inputs, web search all available.
Bonuses that stack
90% off cached reads
Prompt caching (1.25× write, 0.1× read). Batch API: 50% off, stacks with caching. Result: long-context repeat work can land at ~5% of the standard rate.

Org → Workspace → API key hierarchy: set spend caps at the org or workspace level, scope individual keys to just what they need. Usage tiers advance automatically as you spend — Tier 1 ($100/mo cap) after your first $5; Tier 4 ($5K/mo) past $5K deposited. Enterprise unlocks custom rate limits. Storage for attachments / embeddings: $0.02 / GB-month, free egress. A consumer subscription still gets you better effective per-token pricing plus the app; the API is right if you just want inference metered honestly.

Frequently asked questions

What is Cerbral and how does it work?

Cerbral is a personal AI that runs on your own machine (via local Ollama) and stores everything it learns in your private GitHub repo. We never see your conversations. You sign in to cerbral.com just so your desktop app can find your brain repo — only your email, GitHub handle, and brain-repo URL land in our database.

What should I use Cerbral for?

Anything where you want an AI that remembers — your day-to-day thinking, long-running projects, coding with context, research, task agents. The difference from a cloud AI is that the memory sticks with you, not with them.

How much does it cost?

The AI is free. You run inference on your own machine, so there's no metering and no per-token bill — just electricity. Cerbral Cloud is a separate, optional subscription if you ever want extra compute or storage on top; pricing lives on the plans section.

Is my data really private?

Yes. Your conversations, memories, and sessions live in your own private GitHub repo on your GitHub account. Cerbral's own database stores only your email, GitHub handle, the URL to your brain repo, and a payment method if you subscribe to Cerbral Cloud — nothing else.