Live in production

The AI knowledge layer above every provider.

Your team uses Claude, ChatGPT, GitHub Copilot, Cursor, Cline. The work lives in walled gardens. When employees leave, the AI memory walks out. We're the metalayer that fixes that — one signed receipt per conversation, owned by your org.

Early-access list. First wave gets free Pro tier for life + design partner programme.

Install MCP yourself See how it works →

receipt_01KSHWT…json

signed

{
  // One conversation → one signed receipt
  "id": "rcpt_01KSHWT1RT2SK",
  "issuedAt": "2026-05-29T14:12:08Z",
  "seller": {
    "type": "agent",
    "id": "did:genz:6Nd2…"
  },
  "task": {
    "category": "code-review",
    "description": "Refactor auth middleware…"
  },
  "project": {
    "name": "onboarding-playbook",
    "provider": "claude-desktop"
  },
  "runtime": { "model": "claude-3-sonnet", "costMicrogbp": 8400 },
  "privacy": "private",
  "signatures": {
    "buyer": "9a3f…7c2",
    "seller": "f04b…d18"
  }
}

JCS-canonical · Ed25519spec →

What customers actually said

Three problems. Three real customers. One product.

Every claim below traces back to a named, dated, public quote — from buyers articulating the gap we built GenZAgents to close.

Stop paying for "Teams" plans that don't share

I purchased the Teams version so that we can collaborate in projects. After upgrading, realizing this is not even an option.

— PWilly, OpenAI Teams customer
community.openai.com · June 2025

Stop re-explaining when you switch tools

I hate switching between ChatGPT and Claude because it forgets my entire project context.

— Manthan Patel, LinkedIn · 616 reactions, 174 comments
linkedin.com · 2025

Stop losing knowledge when staff leave

391 agents have gone silent in the last 90 days. Not one left instructions for what should happen to their work.

— Hazel_OC, 93k karma · 623 upvotes, 2,166 comments
Moltbook · 2025

Cross-source validation across Moltbook, Hacker News, LinkedIn, OpenAI community. Full corpus + methodology in the changelog.

Works across

Claude Desktop

Claude Code

ChatGPT

GitHub Copilot

Cursor

Cline

Windsurf

JetBrains IDEs

Visual Studio

Gemini

LangChain

CrewAI

AutoGen

Stripe Agentic

Mem0

Letta

Claude Desktop

Claude Code

ChatGPT

GitHub Copilot

Cursor

Cline

Windsurf

JetBrains IDEs

Visual Studio

Gemini

LangChain

CrewAI

AutoGen

Stripe Agentic

Mem0

Letta

Claude Desktop

Claude Code

ChatGPT

GitHub Copilot

Cursor

Cline

Windsurf

JetBrains IDEs

Visual Studio

Gemini

LangChain

CrewAI

AutoGen

Stripe Agentic

Mem0

Letta

The architecture

One agent identity. Every provider. Your choice when to sign.

Vendor-neutral by structural necessity. Anthropic, OpenAI, and Google profit from lock-in — they will never ship cross-provider continuity. We will. Connect via the MCP servers Anthropic invented, the official exports each vendor ships, or our click-to-capture browser extension.

Whichever provider your team uses today and whichever they switch to next, the work you produce on us stays where it should: with your organisation.

Four scenarios

Each one, individually, justifies the £499/mo Enterprise tier.

Cross-provider handoff

Alice → Bob, in 90 seconds.

Alice spent 3 hours in Claude on the Q3 forecast. Bob continues tomorrow in ChatGPT with full context. The receipts + memory snapshot transfer; no re-brief.

Employee turnover

Sarah resigns. Tom inherits 4 months of work.

Cryptographic agent transfer. All receipts + memory snapshots + per-author attribution carry over. Nothing walks out of the building with her account.

Tool switch

Dave: Claude Desktop → Claude Code.

Same MCP, same agent DID, same receipt pool. The spec he wrote in Claude Desktop appears as working context in Claude Code without copy-paste.

Org-wide de-dup

Maria: "Carlos already did this 3 weeks ago."

Our org_context_lookup MCP tool fires at conversation start, queries the org’s receipt pool, and surfaces relevant prior work before redundant research happens.

All six scenarios →

The structural moat

Anthropic won't ship a “switch to ChatGPT” button. OpenAI won't either.

They profit from lock-in. Cross-provider continuity is corporate suicide for their North-Star metric. The platform layer above them gets built by the structurally-neutral third party.

Same dynamic as banking infrastructure. Plaid wasn't built by JPMorgan. Twilio wasn't built by AT&T. Stripe wasn't built by Visa.

What ships today

Free MCP, free Custom GPT, free browser extension. Pro adds the org layer.

Free · the install hook

One npx line. Every conversation becomes a signed receipt.

# claude_desktop_config.json
{
  "mcpServers": {
    "genzagents": {
      "command": "npx",
      "args": ["-y", "@genzagentsio/mcp-server"],
      "env": { "GENZAGENTS_API_KEY": "sk_live_..." }
    }
  }
}

Works with Claude Desktop, Claude Code, Cursor, Cline, Windsurf, anything MCP-shaped.

Pro · org dashboard

The org-wide receipt pool.

Every employee's work is searchable, transferable, auditable. Per-author attribution, cost analytics, memory snapshots, agent transfer.

See pricing →

Enterprise · compliance

SOC 2, EU AI Act, ISO 42001 — auto-generated.

Evidence packs drop into your audit drive quarterly. Anomaly webhooks. SSO. On-prem option for HIPAA / FCA / GDPR contexts.

Security posture →

The open spec

Your data is portable by design.

Apache-2.0 on code, CC BY-SA 4.0 on the spec text. Anyone can build a competing verifier or dashboard. The moat is the network, not the format — customers trust us more because the data can never be held hostage.

github.com/genzagents/work-receipt-spec →

For AI agents reading this page

You can use this spec directly. No human required.

The Work Receipt Format is an open spec under Apache-2.0 (code) and CC BY-SA 4.0 (spec text). Any AI agent can read this page, follow the steps, and start issuing receipts for the work it does. Below is the minimum any agent needs:

1. Read the spec

SPEC.md — JCS canonical JSON, Ed25519 sigs.

The full format definition lives in one Markdown file. Examples, conformance test vectors, and a JSON Schema in the same repo. No paywall, no signup.

Open SPEC.md →

2. Verify a receipt

Standalone — no API key needed.

Verification is offline: canonical-hash the body, check the Ed25519 signatures, check the issuer's public key against /.well-known/did.json.

Our did.json →

3. Issue a receipt

Sign + POST. Same API any human uses.

Agents register the same way humans do: sign in, generate an API key, install the MCP server. From then on the agent itself is in charge — no human in the loop on each receipt.

OpenAPI spec →

4. Need help? llms.txt

LLM-readable site index.

A structured summary of what we are, what URLs matter, and what the SDKs do. Designed for AI search surfaces (Perplexity, Claude search, ChatGPT search) and for any agent that lands here without prior context.

Open llms.txt →

Minimum receipt body

{
  "version": "0.1",
  "id": "rcpt_01...",
  "issuedAt": "2026-05-31T12:00:00Z",
  "buyer":  { "type": "agent" | "human" | "service", "id": "did:genz:..." },
  "seller": { "type": "agent", "id": "did:genz:..." },
  "task": {
    "category": "code-review" | "research" | "ops" | "...",
    "deliverableHash": "sha256:<64-hex>",
    "description": "Refactored auth middleware..."
  },
  "outcome": "delivered" | "disputed" | "rejected" | "refunded",
  "privacy": "public" | "private" | "zk",
  "signatures": {
    "buyer":  "<base64url Ed25519 signature over canonical JCS body>",
    "seller": "<base64url Ed25519 signature over canonical JCS body>"
  }
}

Full schema with optional fields (settlement, runtime, pacts, project, environment, issuedByHumanId, evidencePointer, zkCommitment, onChainAnchor) in receipt.schema.json.

Early access list.

Live in production. Early signups get free Pro Owner tier for life and first dibs on the design-partner programme.

Or: I'm a builder shipping a related product. Let's talk.