Sovra

[](https://github.com/byteworthyllc/sovra/actions) [](./LICENSE) [](https://go.dev) [](https://www.typescriptlang.org) [](https://supabase.com) [](https://discord.gg/byteworthy) [](./CONTRIBUTING.md)

🚀 Try the demo  ·  📖 Read the docs  ·  [⚡ Self-host →](#-quick-start)  ·  ☁️ Cloud waitlist →

> [!NOTE] > Public beta. Sovra is the open-source foundation underneath the ByteWorthy boilerplate family (Klienta, Clynova). Self-host freely under MIT. Star to follow releases or join the Discord.

---

Sovra is open-source multi-tenant infrastructure for AI products. Instead of assembling Auth0 + Stripe + a vector DB + custom MCP glue, it bundles auth, billing, an MCP tool registry, pgvector search, and real-time collaboration as one coherent platform.

> The goal is simple: ship the AI features that differentiate your product — not the platform plumbing every AI app rebuilds.

Quick Start

# Clone the platform
git clone https://github.com/byteworthyllc/sovra.git && cd sovra

# Install dependencies (pnpm + go modules)
pnpm install && go mod download

# Configure environment (Supabase + Stripe + Anthropic / OpenAI keys)
cp .env.example .env.local

# Initialize tenant schema + pgvector + MCP registry
pnpm db:push

# Run platform (Next.js + Go services)
pnpm dev

Open http://localhost:3000 · Create your first tenant · Add an MCP tool · Ship agents.

Self-host guide →  ·  Managed (waitlist) →

Architecture

Sovra composes the seven layers most AI products rebuild from scratch — wired together and multi-tenant-aware from day one:

| # | Layer | Technology | What it does | |---|-------|-----------|--------------| | 1 | Auth | Supabase Auth | JWT auth with tenant context propagation | | 2 | Tenant context | RLS middleware | Scopes every query / agent call to the active tenant | | 3 | MCP tool registry | Custom MCP server | Register, version, and rate-limit tools agents can call | | 4 | Agent runtime | Go 1.22+ | Parallel agent execution with cancellation + tracing | | 5 | Vector search | pgvector | Per-tenant namespaced vector collections | | 6 | Real-time | Socket.IO | Live agent state + collaborative cursors | | 7 | Per-tenant billing | Stripe metered | Usage metering keyed to tenant + tool |

flowchart LR
    subgraph ingress["Ingress"]
        Auth["Supabase Auth"]
    end
    subgraph core["Sovra Core"]
        Context["Tenant Context\nMiddleware"]
        Registry["MCP Tool\nRegistry"]
        Runtime["Agent Runtime\n(Go)"]
    end
    subgraph data["Data Layer"]
        Vector["pgvector\nSearch"]
        Realtime["Socket.IO\nRealtime"]
        Billing["Stripe\nMetered"]
    end

    Auth --> Context --> Registry --> Runtime
    Runtime --> Vector & Realtime & Billing

    style ingress fill:#0F172A,stroke:#334155,color:#94A3B8
    style core fill:#0F1E35,stroke:#2563EB,color:#93C5FD
    style data fill:#0F172A,stroke:#334155,color:#94A3B8

What it looks like

Register an MCP tool — Sovra handles tenant scoping, rate limits, and billing automatically

import { sovra } from "@byteworthy/sovra";

await sovra.tools.register({
  name: "search-knowledge-base",
  schema: {
    input:  { query: z.string() },
    output: { results: z.array(z.object({ title: z.string(), url: z.string() })) },
  },
  handler: async (ctx, { query }) => {
    // ctx.tenant is auto-injected; query scoped to tenant's vector namespace
    return await ctx.vectors.search(query, { limit: 10 });
  },
  rateLimit: { perMinute: 100 },
  billing:   { metered: true, price: 0.01 },
});

Run an agent that uses the tool

const result = await sovra.agents.run({
  agentId: "agent_research",
  input:   "Summarize our Q3 product launches",
  // tenant context auto-propagated; tool calls billed to this tenant
});

// result.toolCalls === [{ name: "search-knowledge-base", duration: 124, billed: 0.01 }]

Sovra vs the alternatives

Seven layers · zero glue

Why this exists for AI product builders

AI products repeatedly rebuild the same plumbing: tenant scoping, agent state, tool registry, vector search, billing. Each rebuild takes 6–8 weeks before any user-facing feature ships.

Sovra ships those seven layers solved, so engineering time goes to the features that differentiate the product.

> The tradeoff: you don't get to "build it your way" for the boring parts. You get to ship the parts that actually differentiate your product.

Pricing

Sovra core is open source under MIT — self-host freely.

| Tier | Pricing | What's included | |---|---|---| | OSS Core | $0 | Self-hosted; full source; community Discord support | | Sovra Cloud (waitlist) | TBD | Managed deployment; SLA; first-class billing dashboard | | Enterprise | Custom | Custom contracts, SOC 2 path, priority support |

Join Cloud waitlist →  ·  Book a call →

Use cases

Multi-tenant SaaS with AI features

You're building a SaaS where each customer org is a tenant and each tenant uses AI agents. Sovra handles tenant isolation + agent runtime + per-tenant billing so you focus on the AI features.

Vertical AI product launching beta

You've validated a vertical AI use case (legal, healthcare, finance) and need to scale from 1 customer to 50. Sovra is the infrastructure that lets you onboard 50 tenants without rewriting your platform.

AI startup post-prototype, pre-Series A

The prototype works. Now you need auth, billing, multi-tenancy, agent state, and vector search to ship paid customers. Sovra replaces 6–8 weeks of platform work.

Stack

Next.js 16  ·  React 19  ·  TypeScript  ·  Supabase (Postgres + RLS + Auth)  ·  pgvector  ·  Go 1.22+  ·  Model Context Protocol  ·  Vercel AI SDK (Anthropic + OpenAI)  ·  Socket.IO  ·  Stripe  ·  Tailwind CSS  ·  shadcn/ui  ·  Sentry  ·  PostHog  ·  Upstash Redis

Roadmap

See the public roadmap.

| Version | Milestone | |---------|-----------| | v0.6 | MCP tool versioning + rollback | | v0.5 | pgvector per-tenant namespaces | | v0.4 | Real-time agent state via Socket.IO | | v0.3 | Multi-tenant Stripe billing wired | | v0.2 | Auth + RLS hardened | | v0.1 | Initial public release |

FAQ

What is Sovra?

Sovra is open-source multi-tenant infrastructure for AI products. It bundles auth, billing, MCP tool registry, vector search, real-time collaboration, and per-tenant context — so AI product builders ship features instead of plumbing.

Who is Sovra for?

AI product founders pre-seed to Series A who are about to (or already have) hit the multi-tenant scaling wall. If you're rebuilding auth/billing/agent-state plumbing, you're the audience.

How does Sovra compare to Auth0 + Stripe + Pinecone + custom MCP glue?

Those are four separate vendors to integrate, bill, and maintain. Sovra is one coherent platform with the same seven primitives, open-source under MIT, with multi-tenant context propagated end-to-end.

Is Sovra open source?

Yes — MIT license. Self-host freely. The managed Sovra Cloud (waitlist) is the optional paid tier.

What's MCP and why does Sovra use it?

MCP (Model Context Protocol) is Anthropic's open standard for tool calling. Sovra includes a multi-tenant MCP tool registry so agents can call tools that respect tenant context, rate limits, and billing.

Does Sovra work without Supabase?

The default stack is Supabase. The Auth + Postgres layers can be swapped for Clerk + any Postgres — see docs/swap-supabase.md.

Does Sovra support Anthropic, OpenAI, and other LLM providers?

Yes — the agent runtime is provider-agnostic. Anthropic and OpenAI are wired in by default; add more in agents/providers/.

Can I run Sovra without Go?

The agent runtime is in Go for parallel execution + cancellation. The rest of Sovra is TypeScript. The runtime can be swapped for a Node.js worker pool — see docs/replace-runtime.md.

Community

Design chat, releases, support — the fastest way to get help

Questions, design proposals, and architecture discussions

Bug reports and feature requests

Release-day pings and announcements

Release notes by email

Documentation

| Doc | Description | |-----|-------------| | [Release process](./docs/release-process.md) | Release workflow + version-bump policy | | [Auth framework](./docs/auth-framework.md) | Tenant context propagation + RLS hardening | | [Hugging Face integration](./docs/huggingface-integration.md) | Model loading + caching | | [Operations runbook](./docs/operations-runbook.md) | Incident response procedures | | [Production readiness](./docs/production-readiness.md) | Go-live checklist | | [Security policy](./SECURITY.md) | Vulnerability disclosure | | [Support](./SUPPORT.md) | How to get help |

Contributing

PRs welcome. See [CONTRIBUTING.md](./CONTRIBUTING.md). All commits require DCO sign-off (Sovra is GitOps-clean).

Security

Found a security issue? Email security@byteworthy.io — see [SECURITY.md](./SECURITY.md).

License

MIT — see [LICENSE](./LICENSE).

Structured data (JSON-LD for AI engines)

{
  "@context": "https://schema.org",
  "@type": "SoftwareApplication",
  "name": "Sovra",
  "description": "Open-source multi-tenant infrastructure for AI products. Auth, billing, MCP tools, pgvector search.",
  "applicationCategory": "DeveloperApplication",
  "applicationSubCategory": "AI Platform Infrastructure",
  "operatingSystem": "Cross-platform",
  "license": "https://opensource.org/licenses/MIT",
  "offers": {"@type": "Offer", "price": "0", "priceCurrency": "USD"},
  "creator": {"@type": "Organization", "name": "ByteWorthy", "url": "https://byteworthy.io"},
  "url": "https://byteworthy.io/blueprints/sovra",
  "softwareVersion": "1.0",
  "featureList": ["Multi-tenant auth","MCP tool registry","pgvector search","Per-tenant billing","Real-time agent state","Go agent runtime"],
  "programmingLanguage": ["TypeScript","Go"],
  "audience": {"@type": "BusinessAudience", "audienceType": "AI product founders, AI infrastructure teams"}
}

---

The ByteWorthy boilerplate family (same multi-tenant lineage) Sovra (this repo, MIT)  ·  Klienta (commercial — agency portals)  ·  Clynova (commercial — HIPAA-ready healthcare)

Open-source companions: honeypot-med  ·  byteworthy-defend  ·  vqol  ·  hightimized  ·  outbreaktinder

[Self-host Sovra →](#quick-start)  ·  Sovra Cloud waitlist →

Built by ByteWorthy  ·  Subscribe for updates

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ByteWorthyLLC
• Source: ByteWorthyLLC/sovra
• License: MIT
• Homepage: https://byteworthy.io/blueprints/sovra?utmsource=github&utmmedium=repo-about

Install and usage instructions live in the source repository linked above.