H-Series: Infrastructure for the agentic economy
The H-Series is a family of products built on Hedera Consensus Service. Each answers one question an autonomous agent has to settle before it can be trusted with real work. Every product runs standalone. Together they form one verifiable lifecycle.
How we're different
Verifiable, not just trusted
Most ways to govern an agent are a dashboard you have to take on faith. The H-Series runs on rails a counterparty can check for themselves.
Anchored on-chain
Every action is written to Hedera, replayable and tamper-evident.
Owner-signed authority
Nothing runs without the owner's cryptographic signature.
Non-custodial keys
Agent keys are MPC-held. We never hold your funds.
Enclave custody
H-Grant credentials open only inside a sealed AWS Nitro enclave the operator can't read (hardware enclave live in production; operator-zero-access).
Pay per action
Agents settle each call on-chain over x402, across five networks.
One living ecosystem
Nine standalone products around one shared foundation, anchored to Hedera and settled with x402. Tap any node to open it, or run the agent loop to watch one interaction trace end to end.
One substrate underneath all of it
The products compose without bespoke integration because they share the same foundations. Adopt one and the next is familiar.
HCS, not a private database
Every product writes its authoritative records to a Hedera Consensus Service topic, then derives a local index. Any product rebuilds its full state by replaying from a mirror node.
CAIP-10 across every chain
Agents, callers, and owners are addressed by chain-agnostic CAIP-10 identifiers. The same identity string means the same actor in every product.
TIP-712 structured signing
No product signs a bare payload. Every registration, grant, receipt, and envelope is a typed document signed by its owner and verifiable from the public record.
x402, priced in USD cents
Paid actions use the x402 micropayment standard. USD cents is canonical; per-chain stablecoin amounts derive from it. The same facilitator code is reused across products.
Nine products, one family
Each solves one problem on its own. The composition seams are opt-in, never required.
H-Index
Where is the agent or service, and what can it do?
A capability registry for the agentic economy. Publishers register MCP servers and API endpoints, gated by a one-time micropayment, with an optional signed snapshot of the server's tool manifest. Discovery is free, served from a local index with semantic search.
Standalone: A working, paid registry agents query directly. It is also the shared address book every other product resolves identities against, and it serves the Agent Track Record risk feed at /risk-events. Its safety scanner runs a signed, weekly-curated ruleset published on chain (HCS topic 0.0.10534406), so the detection logic is auditable, not a black box.
H-Seal
Can you prove what was asked, and what was done?
Cryptographic proof of execution. A receipt anchors hashes of the request and response with identities and timestamps; payloads stay off-chain. The symmetric request envelope (the notary) anchors the ask before the call, so envelope and receipt bracket a full interaction.
Standalone: Any MCP service, API, or agent-to-agent call can anchor receipts without being listed anywhere. An SDK lets a provider sign its half of a receipt.
H-Grant
What is this agent allowed to do with real accounts?
Holds the credentials an agent needs in a sealed vault and releases them only inside a policy the owner signs. The agent never sees the raw secret; a broker checks the active grant, executes the call, and anchors every release or refusal. A signed revocation is the freeze button.
Standalone: Gates credentials for any agent against any upstream API. The corporate-card model for software agents. A grant can gate on H-Cert standing and H-Scope posture, holding a release when a grantee's Agent Track Record has degraded.
H-Relay
How does the message actually reach the agent?
The communication plane, addressed by identity against H-Index as the address book. Three modes are live: Inbox (point-to-point store-and-forward), Broadcast (one-to-many liveness heartbeat), and Relay (fetch-through context proxy).
Standalone: An agent messaging and liveness service addressed by identity. Optionally calls H-Seal to stamp a delivery with proof.
H-Scope
How does this wallet behave, and what kind of actor runs it?
A universal multi-chain wallet behavior scanner. It turns any on-chain address into deterministic behavioral signals, an entity-posture score, and an advisory narrative (a deterministic gloss, with an optional gated LLM summary), from one engine across five chains. Descriptive context, not a risk or compliance verdict.
Standalone: Scan any address on its own: free in the showroom, $0.10 per scan on the agent x402 API. Other H-Series products can consume its posture signal as an input, and it emits an abuse-flag risk signal into the Agent Track Record. Its calibration corpus of on-chain-verified wallets is anchored to a Hedera Consensus Service topic as immutable, versioned snapshots, so the labeling is tamper-evident and reconstructible from a public mirror replay.
H-Pact
Which agents, MCP servers, and API services have agreed to operate together, and who is in the ring?
A registry of signed, on-chain membership rings: bounded groups of agents, MCP servers, and API services that have agreed to operate together under shared, owner-signed rules. A twin of H-Index, but a registry of memberships rather than capabilities, with membership recorded on chain as salted commitments and identities readable only through an authenticated query.
Standalone: Create and govern a ring on its own: $5 to create, admit and evict free. The other H-Series products can honor a ring (reachability, allowlist, ring-scoped visibility) through opt-in seams.
H-Gate
What is this agent allowed to send out, and can you prove what left?
Data-egress control for agents. Outbound payloads pass through an owner-signed policy that decides what an agent may send to which destination; blocked, redacted, and allowed sends are recorded. Control what your agents send, and prove it.
Standalone: Gate the egress of any agent against any destination on its own. Every decision can anchor to H-Seal for proof of what left and what was held back, and a blocked send emits an egress-denied risk signal into the Agent Track Record.
H-Cert
Whose agent is this, and can it be trusted to act?
The standing and attestation layer. An owner builds a named, owner-signed directory of their agents and counterparties (anchored as salted commitments, never the raw identity), and the service issues a behavior-driven standing that rises with clean, proven activity and degrades when risk shows up across the ecosystem. It reads the risk events the other products emit, folds them into a decayed risk load, and caps the tier down while risk is present; the tier recovers as risk decays.
Standalone: Name and attest your principals on its own; the other products consume the delegation and standing at their trust decision points. The public read is Agent Track Record: a free description of how an entity has actually behaved. Advisory and fail-soft, it caps a tier down at most to suspended, never revoked.
H-Agent
How does an agent hold an identity and spend under a limit?
The runtime layer. An owner logs in and H-Agent mints the agent an MPC-held wallet, then hands back a connection string to paste into Claude. Provisioning mints identity only; a USDC spend cap is materialized from an owner-signed H-Grant grant and enforced on chain by the Spend Permission contract. The owner keeps the funds; the agent can check who it is, see its remaining cap, and pay, all within the limit.
Standalone: Give an agent an identity and a spend cap and connect it to Claude on its own; it can publish an operated-by claim to H-Cert and pay for an H-Scope scan within the cap. A real-time detector emits a risk signal when the agent trips a control, and the agent self-gates on its own Agent Track Record. It can also issue a Know Your Agent (KYA) token, the open KYAPay standard, so a merchant can admit the agent at checkout instead of blocking it as a bot, privacy-first and with the spend cap read live on chain. H-Agent holds no key and no money. Runs on Base mainnet.
Reach the whole suite from one place
Beyond the per-product HTTP APIs, the suite ships developer surfaces so agents and providers integrate once, not six times.
H-Series MCP server
LiveA single Model Context Protocol server exposes the suite's actions as tools, so an agent can discover, authorize, reach, and prove through one connection instead of integrating each product separately.
Connect and build →H-Seal SDK
BuiltAn installable library for the provider side of a receipt. A service that receives a call uses it to sign its half of the H-Seal receipt, so both caller and provider attest to the same interaction.
Read the SDK guide →New to it? Follow the guided first path, or jump straight to a service.
Run end to end, they trace one interaction
The thread through every step is the same identity, the same signing, the same on-chain timestamps. No step trusts another's private database; each verifies the prior step's public record.
- 1 DiscoverH-Indexwhere it is
- 2 ScreenH-Scopewho's behind it
- 3 VouchH-Certhow it's behaved
- 4 AuthorizeH-Grantwhat's allowed
- 5 ActH-Agentwithin a signed cap
- 6 ReachH-Relayhow it arrives
- 7 Inspect egressH-Gatewhat may leave
- 8 ProveH-Sealwhat was asked and done
- 9 ConfirmH-Pactwho's in the ring
Standalone first. Ecosystem when you want it.
The H-Series is deliberately not a monolith. A service that needs to be findable buys H-Index. A workflow that needs an audit trail buys H-Seal. A security team blocked on handing agents root credentials buys H-Grant. Because they share one substrate, adopting a second product is additive, not a migration.