Pick a tier · price it · hook up

Assurance Tiers

Four tiers.
Each one says exactly what it does.

You should never have to choose a proof. You just choose how much assurance your task needs. Each tier is named plainly, so there is nothing to decode. Pick a tier below and the price, the savings estimate, and the hook-up button all follow it. Verification is always free; you only pay to sign. Each tier maps to the AI receipts proof layer and its AI audit trail.

verify: true One flag, plus a tier. That is the entire integration. Your agents request the tier their task needs; Hive applies the right proofs. No primitive menu, nothing to decode.

The ladder: tap a tier to select it

Selected

Verified everyday inference

Every answer is signed with cryptography and grounded in its sources. Anyone can check it, offline, with no key.

What it covers

High-volume answers where you want proof without ceremony, like RAG, search, or chat
A signed receipt per answer, plus per-claim grounding to its sources

Choose if

You want the meter to spin and every answer to be provable.

Per receiptfirst 1M free / mo

Then$0.0008 / receipt · Scale

License from$120k / yr · single product

Example10M answers/mo ≈ $7,200/mo

Start with Verified Estimate my cost Mint a free receipt

Selected

Verified Model multi-model · routing · fine-tunes

And you know exactly which model produced it. It is signed, grounded, and provably from the model you think it is. No silent swap, no hidden lineage.

Adds to Verified

Proof of which model answered and what it descends from, the DeepSeek / Kimi / Qwen question
A model-identity receipt: which model, derived from what, attested how

Choose if

You serve multiple models, route, or fine-tune open weights and must prove what ran.

Per receipt$0.0008 → $0.0004 · Scale to Platform

License from$120k / yr · single product

Example100M answers/mo ≈ $40k/mo, or license

Start with Verified Model Estimate my cost Mint a free receipt

Selected

Verified Agent agentic products

…and every move your agent makes is provable. Signed, grounded, model-verified, and a tamper-evident record of everything your agent did, observed, altered, and was authorized to do.

Adds to Verified Model

The agentic stack: every tool call, context change, and action, signed
The state the agent saw is signed before it acts, not logged after the fact
Routing by proven behavior
AFiR-RC · recoverable checkpointing: the agent catches a bad step, rolls back to the last good signed checkpoint, repairs itself, and continues, with every rollback recorded in the chain. Patent Pending.
AFiR-Stream · real-time streams (voice agents, live translation, streaming copilots) get attested segment by segment off-path, with zero added latency, and the whole session can be checked offline. Patent Pending.

Choose if

You ship agents that take real actions and need to prove they behaved.

Per receipt$0.0004 → $0.0002 · Platform to Hyperscale

License from$360k / yr · full suite

Examplean agent run means many receipts, so a license usually wins

Start with Verified Agent Estimate my cost Talk to us

Selected

Verified Regulated banks · health · legal · gov · labs

And it is audit-grade, hardware-attested, and it catches instability. You get everything above, plus hardware-attested proof, instability detection, and provable evaluation with contamination proof. Nobody else can sell this kind of assurance at any price.

Adds to Verified Agent

Hardware-attested signing on a hardware root
Instability detection and triangulation
Provable evaluation with contamination proof
AFiR-Stream · compliance boundary · per-utterance proof for real-time voice (PCI, HIPAA, PHI). Each segment carries a signed record of sensitive-data class, the model that handled it, and whether it stayed inside the compliance boundary. Patent Pending.

Choose if

A wrong, unprovable, or unstable answer is a legal or existential event.

Per receiptHyperscale rate + premium add-ons

LicenseSuite $360k / yr + sovereign / on-prem

Priced againstthe cost of a regulatory finding, not inference

Start with Verified Regulated Estimate my cost Talk to us

Each tier is a superset of the one below. More assurance is always just the next tier up. You never have to re-integrate.

Pick your world, see your tier

Every vertical has a natural floor, the lowest tier that clears its assurance bar. Tap a tier to select it above.

Your world	Floor tier	Why this tier	The pain it removes
Search / RAG / chat	VERIFIED	High volume, low ceremony: proof as a feature	Is this answer grounded or hallucinated? Every answer carries its sources, signed.
Multi-model / routing / open weights	VERIFIED MODEL	Must prove which model actually ran	Silent model swap, hidden lineage, the model-identity question.
Agentic products that act	VERIFIED AGENT	Every action provable, not logged after the fact	Reconstructing what the agent saw takes weeks; here it is signed before the act.
Healthcare	VERIFIED REGULATED	Audit trail must survive long retention; clinical inference is high-consequence	Unprovable clinical-decision-support inference; audit trails that rot.
Finance	VERIFIED REGULATED	Tamper-evidence, enforceable SLAs, trading and credit decisions	"Show your work" to a regulator; disputed bills; unenforceable latency language.
Legal	VERIFIED REGULATED	Chain-of-custody on e-discovery and contract-analysis inference	Repudiable timestamps; privilege boundaries with no proof.
Government / Defense	VERIFIED REGULATED	Post-quantum-mandated signing on a hardware root	Classical crypto that will not clear a PQC mandate.
Motorsport	VERIFIED AGENT to VERIFIED REGULATED	AI runs the pit limiter, brake-by-wire, and energy map, and incidents need to be reconstructable	Weeks-long incident reconstruction with no signed record of the state the AI saw.

Price it: drag the dials

Verification is always free. You only pay to sign. Set your volume and what a failure costs you; the math is live.

Your tier

Signed receipts per month 10,000,000

10K100K1M10M100M1B10B+

What one unprovable failure costs you $50,000

$1K$10K$100K$1M$10M+

Failures you avoid per month 2

0per month50

Your Hive cost $7,200 / mo 10M receipts at the Scale rate, first 1M free

What it protects

$100,000 / mo

Net upside

$92,800 / mo

At this volume, the metered Scale rate is your cheapest path. A single-product license becomes cheaper above roughly 12.5M receipts/mo.

Start with Verified Talk through the numbers

This is an estimate only. Your real invoice is metered per signed receipt and settled in USDC on Base. Verification of any receipt is always free. License "from" prices are floors, not exclusivity.

How the choice happens: three easy ways

The human picks a tier

Four names, four promises, five seconds. Each tier is a superset of the one below. More assurance is always the next tier up.

The agent picks by intent

Your agents declare the outcome they need: standard, model, agent, or regulated. Hive maps it to a tier. They ask for the outcome, and we assemble the proofs.

You hook in and Hive auto-selects

One flag, no tier specified. Hive picks by the shape of the call. A RAG answer gets Verified, an agent run gets Verified Agent. Zero configuration still gets you the right proofs.

# your agents request assurance by intent, never by primitive
assurance: "standard"   -> Verified
assurance: "model"      -> Verified Model
assurance: "agent"      -> Verified Agent      (auto when the call is an agent run)
assurance: "regulated"  -> Verified Regulated

The integration

Add one flag:          verify: true   (+ optional tier, or let it auto-select)
Expose to the sidecar:  { query, sources, answer, model-id }   (out-of-band, no latency on the answer)
Attach receipt_id to the response object.

That's it. No primitive selection. No re-architecture. Fails open.

You sell "verified inference" as a feature; Hive is the engine. You flip one flag, your agents ask for the assurance they need, and we handle which proofs apply.

Same speed. Now provable.

Verification is not a tax on your answer. The proof is minted out-of-band, in parallel, and attached after the fact. Your tokens ship at the speed they do today. We did not make inference slower to make it provable.

Latency on your answer

0 ms added

Signing runs alongside the response on a sidecar, not in the token path. It fails open, so if the signer is down, the answer still ships.

Per-receipt signature

~7 ms

One ML-DSA-65 post-quantum signature (NIST FIPS 204) per receipt, off the critical path.

Receipt assembly

53 µs

This is an optimized typed-fragment path, measured 35% faster than the raw-byte baseline at typical fragment counts. It is slower only at very large runs (512+ fragments).

End-to-end signed inference

126 to 322 ms

Critical-path latency across a 10-run distributed benchmark, every run signed and verified. 10 of 10 succeeded.

These are honest numbers. Receipt-assembly medians come from 500 iterations; end-to-end numbers come from a 10-run distributed harness. The signature itself is post-quantum (ML-DSA-65). The point is not that signing is fast on its own. It is that signing never sits on the answer's critical path.

Verified vs. the analog

This isn't Hive versus a faster model. It is the same inference you ship today. A trust-me dashboard claim turns into a fact anyone can recompute and check.

	Ordinary inference	Hive Verified inference
Latency on the answer	baseline	same, signing is out-of-band
The bill	trust the dashboard	a recomputable fact, signed per receipt
The SLA	unenforceable language	“pay-if-slow,” machine-settled from signed measurements
Which model ran	trust the label	proven, no silent swap, no hidden lineage
What the agent did	logged after the fact	signed before the act, tamper-evident
Holds up to a regulator	no	audit-grade, PQC, hardware-rooted, long-retention
Anyone can verify it	no key, no proof	yes, offline, no secret, no account

And if you are wondering who else does this: no one ships it in production. The nearest comparable is a single solo project on a test network using classical, non-quantum-safe crypto with no hardware root. Hive is the only post-quantum, hardware-rooted, publicly-verifiable per-inference receipt running on real rails.

Pricing rides below your inference cost

Verification is always free, forever. You only pay to sign, and the fee falls as you scale.

Tier	Meter (per signed receipt)	License path
VERIFIED	Build first 1M free → Scale $0.0008 → Platform $0.0004	Single-product license from $120k/yr
VERIFIED MODEL	Scale $0.0008 → Platform $0.0004	Single-product license from $120k/yr
VERIFIED AGENT	Platform $0.0004 to Hyperscale $0.0002. The meter spins fastest here since an agent run is many receipts	Suite license from $360k/yr
VERIFIED REGULATED	Hyperscale $0.0002 plus premium add-ons, priced against the cost of a regulatory finding, not against inference	Suite from $360k/yr plus sovereign / on-prem (custom)

Patent Pending across the suite. ML-DSA-65 signatures (NIST FIPS 204). Receipts anchored to Base mainnet with USDC. The full pricing breakdown, both pay-as-you-go and license, lives on the pricing page.

The one line

You don't pick proofs. You pick how much assurance your task needs: Verified, Verified Model, Verified Agent, or Verified Regulated. Then you flip one flag. Your agents request the assurance they need, and Hive handles the rest.

Your move

Every page in Hive uses the same four verbs, so you are never more than one click from acting. Pick up where you left off.

See full pricing Estimate my cost Mint a free receipt Talk to us

Four tiers.Each one says exactly what it does.

The ladder: tap a tier to select it

Pick your world, see your tier

Price it: drag the dials

How the choice happens: three easy ways

The integration

Same speed. Now provable.

Verified vs. the analog

Pricing rides below your inference cost

The one line

Your move

This work is independently verified by Hive

Four tiers.
Each one says exactly what it does.