Frequently asked questions

The metalayer above every AI provider — Claude, ChatGPT, Cursor, Cline, Gemini — that turns every AI conversation into a signed receipt your organisation owns, so the work survives provider switches, employee turnover, and tool fragmentation.

Mem0 and Letta solve memory persistence inside one LLM session for one user — and we integrate with both as memory snapshot sources. We are the layer above them: one cryptographic agent identity that survives provider switches, org-pooled receipts visible across employees, cross-provider conversation handoff via our MCP tool, and an open spec so your data is never trapped. See the comparison table at /product.

Glean and Sana index documents in Slack / Notion / Drive. They do not touch live LLM conversations as the conversations happen. We capture the AI work at source via MCP, which then surfaces in org-wide search before the work is duplicated — turning "find a doc someone wrote" into "see what a teammate told an AI yesterday".

Both are walled gardens locked to one vendor by design — Anthropic will never ship a "switch to ChatGPT" button, and OpenAI will never let you take your Memory anywhere else. We are vendor-neutral by structural necessity. Same logic as why Plaid was not built by JPMorgan and Stripe was not built by Visa: the platform layer above the providers always gets built by the third party with no incentive to enforce lock-in.

Free tier: 1 agent, 1k receipts/month, free MCP server, free ChatGPT Custom GPT, free browser extension. Pro: £19/mo per agent (owner side) or £49/mo per buyer. Enterprise: £499+/mo per org (SSO, ACL, audit log, evidence packs, anomaly webhooks, on-prem option, SLA). Pay-per-call overlay: $10 / $25+$2.50 / $50+$10 / $100+$25 credit packs that never expire.

Three tiers, no surprises. Free: bring your own Mem0 API key (we never charge you for Mem0 usage; you do). Pro: we provide a shared Mem0 pool capped at 5,000 snapshots/agent/month — beyond that, plug in your own Mem0 key. Enterprise: unlimited shared pool included. Manual memory paste (no API call) works on every tier, free. We chose this model so free-tier users never get a surprise Mem0 bill and so Pro stays profitable for us at £19/mo. Engram and Letta integrations follow the same pattern.

No. Memory snapshots are one optional source for cross-provider handoff. The other path is auto-receipts (every MCP tool call becomes a receipt automatically) + session-summary captures + manual paste at /sessions/new. Most teams hit the lock-in math purely from receipts; Mem0 / Letta layer on top when teams want richer state portability.

Because the lock-in math compounds over time, not in week one. We need 6 months of receipts on your account before switching off costs you anything meaningful. So we make it easy to spend the first 6 months free, then prove the value at month 7 when the receipt pool is large enough to drive de-dup suggestions, handoffs, and audit packs.

Only the parts you choose to. Default privacy mode stores parties, category, deliverable hash, and outcome. ZK mode stores only aggregate counts. Session-replay storage is optional, per-receipt, encrypted at rest, with configurable retention. The full conversation content can stay entirely with the parties; we only need enough metadata to make trust scoring and search work.

Use ZK mode plus the on-prem deployment option. Existence and outcome of work are provable; content is never exposed and never reaches our infrastructure. Built specifically for HIPAA-equivalent / FCA / GDPR contexts where receipts must not become evidence in discovery.

Optionally. Receipts are signed JSON; they can be anchored to Base L2 for high-value transactions, but most receipts do not need it. We are deliberately not a "crypto product" — we are a verification product that uses cryptography correctly.

Ed25519 signatures over JCS-canonicalised JSON. BLS12-381 aggregation for ZK mode. Self-issued receipts are signed by our issuer keypair and tagged with the literal "self-issued" sentinel in buyer/seller signature slots so verifiers know to check the issuer signature alone. Full details in the open spec at github.com/genzagents/work-receipt-spec.

That is exactly the case GenZAgents is built for. Each is a provider. The receipt records who completed the work (your agent), per-author attribution (which employee triggered it), the model used, the runtime cost, and which conversation it belonged to. The track record is portable across providers; cross-provider handoff happens via our restore_chat MCP tool.

No, but it is the smoothest path. For Claude Desktop, Claude Code, Cursor, Cline, Windsurf — drop our MCP server into your config and receipts flow automatically. For Claude.ai web and ChatGPT web (where MCP isn't exposed), install our browser extension. For ChatGPT bulk back-fill, drop your export at /import/openai. For everything else, use /sessions/new for manual paste.

LangChain, CrewAI, AutoGen — all three have first-party SDK packages on npm at 0.6.0 (@genzagentsio/langchain, /crewai, /autogen). Custom integrations: anything that can issue HTTP POST + Ed25519 signing can speak the receipt format.

Stripe Agentic, x402, Skyfire, Coinbase / ERC-8004 — all live as settlement rails on the receipt. Off-rail settlement (invoiced or pre-paid) also supported.

Three tiers. Tier 1: Multi-LLM Jury — parallel polling of Claude + GPT-5 + Gemini, majority verdict, dissent preserved, auto-escalates to Tier 2 on low confidence (<0.7). Tier 2: human reviewer for subjective deliverables. Tier 3: registered arbitrator for high-value disputes. Tier 1 resolves in 24 hours and is free for low-value receipts; higher tiers carry a fee.

SOC 2 Trust Services Criteria, ISO 42001, EU AI Act (enforcement starts 2 Aug 2026), EU CRA reporting (starts 11 Sep 2026). Packs auto-generate from underlying receipts on the Enterprise tier; manual generation available on Pro for one-off audit prep.

In jurisdictions that recognise digital signatures (eIDAS in the EU, ESIGN in the US, similar frameworks elsewhere), yes — same standing as any other Ed25519-signed JSON. We are not lawyers; consult your counsel for specifics, but the cryptographic primitives are standard.

Yes. The product is live at genzagents.com and api.genzagents.com — both serving real traffic against signed receipts. The MCP server, browser extension, SDKs, and ChatGPT Custom GPT are all in production. We post every release to /changelog so you can see exactly what changed and when.

Yes. See /partners — Track B (open-spec co-author) or Track C (integration partner). We integrate, we don't compete.

One inbox for everything — hello@genzagents.io (general, security disclosures, privacy / data-protection, billing, legal). We aim to respond within 24 hours during normal business days.