Intent router with MEV‑aware settlement
We replaced a legacy order‑router with a solver‑driven intent system. Same UX, 23% better fills.
BlockZenith is an R&D and product house building at the edge of frontier finance and agentic AI. We design DeFi protocols, ship DeFAI agents that execute policy on‑chain, and build LLM systems that actually stay in production.
Our labs ship both as products and as engagements. Everything we deliver is audit‑ready, replayable, and built to outlast the hype cycle.
Bitcoin, Ethereum, Polkadot, and L2 rollups — core code, pallets, precompiles, bridges.
Audited contracts, autonomous vaults, MEV‑aware execution, guardian circuits.
Multi‑agent systems that stay aligned, in budget, and on task — across long horizons.
HSM‑backed signers, attested builds, SLSA‑3 provenance, eBPF observability.
Each lab ships open‑source primitives and an internal product. They share tooling, infra, and post‑mortems.
Four practices. Fixed‑price week‑zero scoping. Long engagements with small teams.
We write these on the office wall and the commit hook. They quietly shape every decision we make, from protocol choices to agent policies.
Every contract, signer path and agent policy ships with attestations.
If a commit hash cannot build the same artifact twice, we do not ship it.
Client data never traverses a third‑party inference tenancy.
Every autonomous decision can be re‑run offline with the same output.
Operators can pause, freeze, or rewind an agent at any block.
We open the parts the market deserves; we sell judgment, not secrets.
We replaced a legacy order‑router with a solver‑driven intent system. Same UX, 23% better fills.
Self‑custodial vaults that rebalance autonomously. Guardian circuits suspend strategies on regime change in under 1 block.
A role‑typed agent graph ingesting tickets, graph data and ledger state. Audit trail is the product.
Authored a production pallet and XCM extension enabling permissioned RWA transfers across the relay.
A deterministic fuzzing harness targeting Bitcoin Core script interpreter. Upstreamed.
FHE‑backed inference so signals never leave the enclave. Latency budget met at production scale.
Taste, ownership, and calm. In that order. These shape our hiring, our scoping, and our Saturday mornings.
We resist fashion for its own sake. If a tool adds complexity, it needs to pay rent in clarity, safety or speed.
From chain to chat, our people read the full code path. No hand‑offs to a black box. No magic.
We prefer a commit over a conference deck. Our public proof lives on GitHub, on‑chain, and in post‑mortems.
If a human is doing the same thing twice, we automate it and give the time back. That is the whole thesis.
Crypto cycles and AI hype cycles both favor the impatient. We plan in years and review in weeks.
Markets break, models drift, incidents happen. We answer with checklists, logs and quiet execution.
A practical migration plan for HSM‑backed signers moving to lattice‑based schemes without breaking end users.
Why we record every tool call, memory write and policy update — and how that record becomes a product feature.
Mapping MEV, oracle drift, protocol drift, policy drift and agent drift into a unified loss framework.
A protocol you cannot risk getting wrong. An agent system that needs to behave. A trading stack that has to be audit‑ready by Q4. We scope in a week and ship in months, not quarters.