Beyond - An Agentic-Platform-First Build Plan for Univrs

An agentic-platform-first build plan for Univrs. Three matrices map layer ownership, design philosophy, and rivals' shortcomings into openings; five inversions define what 'beyond' means; concrete repos, narrow open hardware, and a government-mission services track; five gates from G1 verifiable determinism to G5 a civic service running on the stack.

#univrs#build-plan#agentic-platform#open-source#open-hardware#sovereignty#vudo#strategy

A companion to the platform vision in Universal Nexuspolis and the commons argument in The New Commons and The Grid Commons.

Working document — Sepahsalar Labs / Univrs.io. Comparison matrices, gap analysis, and a sequenced build list.


1. Matrix A — Layer ownership

Who controls what, and on what terms. OSS = public source. SAP = source-available on purchase. Blob = opaque dependency. = not addressed.

LayerSerenum (Smith)OxideUnivrs todayUnivrs target
SiliconBlob (commodity RISC-V SoC)Blob (AMD Turin/Milan + PSP)Blob (4-cloud x86)Blob, but attested and declared
Root of trust / firmware— (no RoT story)Own RoT + SP, Hubris, OSSOpen-hardware RoT node (OSS + OSHW)
BootOwn, SAPOwn, OSS (no BIOS, no BMC)Cloud-vendorMeasured boot → attestation receipt
KernelOwn, SAPHubris (embedded) + illumos/Helios, OSSVUDO as supervisor, not kernel
RuntimeNative RISC-Vbhyve/Propolis VMsVUDO WASM (partly public)VUDO, fully OSS, frozen ABI
Language / compilerBrevis, self-hosted, SAPRust + LLVM (upstream)DOL v0.8.x, self-hosted, OSSDOL as policy surface, not just spec language
Executable formatPB, ownELF + own image formatsWASM + manifestsContent-addressed Spirit, reproducible
NetworkOwn stack, Wi-FiMaghemite, OPTE, P4/Tofino, OSSlibp2p + LoRa + SCION trackMultipath: IP / SCION / LoRa, degrade-not-fail
Storage / stateOwn filesystemCrucible (replicated block), OSSMYCELIUM-SYNC CRDTsCRDTs as DOL language primitives
Control plane— (single user)Omicron, OSSNexus (4-cloud)Nexus + capability grants as the unit
IdentityRack-scopedEd25519Portable sovereign identity + delegation
EconomicsPurchase priceEnterprise saleENR, mutual creditMetering in the scheduler, not the invoice
GovernanceOne personRFD process (public)Constitutional invariants (design)Public RFDs + protocol-level invariants
HardwareProduct, closed designProduct, closed designOSHW edge node (KiCad + BOM)
License postureSAP with deviceMPL-2.0 software, closed HWMixed / some privateOSS-by-default, OSHW where trust is decided
Capital modelOne person, product sales$344M VC, enterpriseSelf-funded + intended grantsMission grants + commons services

2. Matrix B — Design philosophy

DimensionSerenumOxideUnivrs (proposed)
Unit of ownershipThe machineThe rackThe capability grant
Unit of failureOne person’s attentionOne rack, one update domainOne sovereignty domain
Complexity budget~50k LOC, hard ceilingWhatever 120 engineers can operateExplicit LOC + surface ceiling per layer
What it refusesC, POSIX, ELF, LLVM, driversOpaque layers, BMCs, vendor blobsUnattributable action
The unownable blobThe SoCAMD PSP, SSD firmwareModel weights
Verification storyOne mind can read itDTrace, Humility, oximeter, RFDsDeterministic replay of agent execution
Update semanticsWrite once, run foreverSingle rack-wide update domainFrozen ABI + content-addressed builds
Non-determinismNot a design concernNot a design concernFirst-class design concern
Who it’s forThe individual programmerThe enterprise operatorThe polity
RevenueDevice salesRack salesMission services + commons credit

The last two rows are where Univrs is not competing — it is doing something neither of them attempted.


3. Matrix C — Their shortcomings → your openings

ShortcomingWhoseWhy it mattersUnivrs opening
No story for stochastic actorsBothNeither system was designed for a component that can decide. Every safety property assumes deterministic code.Capability gates compiled from DOL specs; authority declared before behaviour
Openness is partial and invertedBothOxide: open software, closed hardware. Serenum: closed repos, source in the box. Neither is reproducible by a third party.Reproducible builds + OSHW for the trust-deciding parts
Single-operator or single-vendor governanceBothSmith is a bus factor of one. Oxide is a cap table.Charter + public RFDs + protocol-level invariants that survive the founder
No multi-generational guaranteeOxideRack lifecycle is 5–7 years and vendor-dependentABI freeze + content addressing as an explicit 30-year promise
No collapse toleranceBothBoth assume grid power, supply chains, and IP transitLoRa mesh + local inference + CRDT convergence after partition
Economics live outside the machineBothMetering is billing, bolted onENR metering as a scheduler input
No public design recordSerenumKnowledge dies with the blogCopy Oxide’s RFD process verbatim — it is free and it is their most transferable asset
Model provenanceBoth (unaddressed)The blob of this decade is weights, not firmwareModel attestation + local-first inference + declared degradation path

4. The five inversions (what “beyond” actually means)

  1. The unit of sovereignty is the capability grant, not the box. Smith owns a machine; Oxide owns a rack; you own the right to act. Hardware becomes an implementation detail of an attestation claim. This reframes every build decision below.

  2. The compiler is the policy kernel. DOL’s “systems describe what they ARE before what they DO” is already the right shape for this. Extend it so that a Spirit’s declared traits and constraints compile into the WASM host-function surface it is permitted to touch. An agent cannot exceed its ontology because the ontology is the ABI. Neither Brevis nor Hubris needed this; you do.

  3. Replay is the new debugger. Oxide’s real moat is not Rust — it is Humility, DTrace, oximeter, and the RFD archive: the ability to explain what happened. For stochastic systems the analogue is deterministic replay. WASM makes this tractable: record every host-function call and return, and any Spirit-plus-agent execution can be replayed bit-exact for audit. Ship this early; it is a credibility artifact for government buyers and a research contribution simultaneously.

  4. Weights are the blob — so build the seam, not the purity claim. Oxide is honest that the PSP remains opaque. Be equally honest: frontier weights are not yours. What you can own is the seam: model attestation, capability-scoped inference, a local-first fallback (the Qwen3/llama.cpp stack you already run), and a declared degradation ladder from frontier → local → deterministic.

  5. Persistence is a protocol property, not a promise. Serenum’s “write once, run forever” is a marketing claim backed by one person’s discipline. Make it a mechanism: frozen Spirit ABI, versioned DOL spec with a conformance suite, content-addressed artifacts, reproducible toolchain. Then a 2026 Spirit demonstrably runs on a 2056 runtime.


5. What to build — concrete

5.1 Open source (new repos)

RepoPurposePrecedent
univrs/rfdPublic, numbered design record with a real review lifecycleOxide RFDs
univrs/spirit-abiFrozen ABI spec + conformance test suite + versioning policyWASI, POSIX
univrs/oracle (replay)Record/replay debugger for Spirit + agent executionHumility, rr
univrs/attestMeasured boot, SBOM, model provenance, attestation receiptsin-toto, SLSA, Sigstore
univrs/capabilityCapability grant format, delegation, revocation, expirymacaroons, UCAN, ZCAP
univrs/conformanceRuns the ABI + capability + CRDT convergence suites against any implementationWPT
univrs/gov-controlsNIST 800-53 / 800-171 / AI RMF crosswalks, SSP templates, as OSSOpenControl

Everything else already exists in some form (dol, vudo, enr, network, nexus). The discipline is: no new capability repos until these seven exist.

5.2 Open hardware — narrow, not heroic

Do not attempt a sled. Oxide spent nine figures and still ships vendor firmware. Build open hardware only where trust is decided:

  • Node R0 — sovereign edge node. RISC-V SoC (SiFive/StarFive/Milk-V class) + LoRa radio (Meshtastic-compatible) + open secure element (OpenTitan-derived or LPC55-class) + measured boot into VUDO. KiCad sources, published BOM, CERN-OHL-S licence.
  • Deliverables that make it real: schematics, gerbers, BOM with second sources, a build guide someone else can follow, and a conformance-passing reference image.
  • Why this and not a server: it is the artifact that makes “sovereignty” checkable rather than rhetorical, it is fundable at grant scale, and it is the piece that survives infrastructure failure.

5.3 Services — government mission track

Sequence it as capability → credential → contract:

  1. Beachhead deliverable. Pick one civic service where the infrastructure claim is visible and the outcome is measurable. (Your Direct File fork is the obvious candidate — verify its current federal status before you anchor a narrative on it, since the programme’s position has moved.)
  2. Compliance as open source. Publish the control mappings, SSP templates, and SBOM/attestation pipeline. This is unusual, cheap, and it converts your Cargill FedRAMP/CMMC experience into a public good and a moat simultaneously.
  3. Entry paths worth pursuing in parallel: SBIR/STTR (DoD, DHS, NSF), DIU solicitations, state digital-service contracts (lower barrier, faster), and research grants where the replay/attestation work is the deliverable.
  4. Positioning. Sell verifiable sovereignty: reproducible builds, attested execution, replayable agent decisions, and continued operation under partition. That is a procurement-legible sentence. “Post-liberal cultural framework” is not.

6. The cultural layer — how to make it load-bearing

You already hold the right instinct: resistance must be encoded at the protocol level, not the policy level. Make that concrete.

PrincipleProtocol mechanism
Subsidiarity (Swiss federalism, satrapy)Sovereignty domains with local-first authority; upward delegation is explicit, scoped, and revocable
Ostrom: clear boundariesCapability grants with named holders and expiry
Ostrom: graduated sanctionsEntropy tax / credit degradation rather than binary exclusion
Ostrom: nested enterprisesDomain hierarchy in Nexus, with no root authority
Multi-generational persistenceABI freeze + conformance suite + content addressing
Narrative as institutional technologyLion & Swan canon versioned alongside the spec, in the same repo
Refusal of absorptionNo privileged operator role anywhere in the protocol — if you can be compelled, so can the system

One piece of honest advice on framing. “Post-liberal” is a contested label that reads very differently to a grant reviewer, an academic, and a contracting officer, and it will cost you in at least two of those rooms. Keep the philosophical work in the Sepahsalar publication track, where the argument can be made properly. In the infrastructure and procurement track, use the operational vocabulary — subsidiarity, commons governance, digital sovereignty, continuity of operations. Same commitments, no unforced translation loss.


7. Honest constraints

  • Surface area is your Serenum problem. DOL, VUDO, ENR, Nexus, SCION, LoRa, Cogito/JEPA, Imaginarium, Direct File, plus the publication track. Smith’s 50k-line ceiling is the real lesson, and it applies to fronts, not just lines. Freeze at least three of those before starting the seven repos above.
  • Self-hosting is not the milestone people think it is. DOL self-hosting is genuine, but Brevis self-hosted in 2022 and Serenum still has no third-party users. The milestone that matters is someone else’s Spirit, built from your published spec, passing your conformance suite.
  • Serenum’s last release was January 2025. One person, twenty months, no release. The failure mode is not technical.
  • Open hardware is a three-year commitment minimum once you have a board that boots. Budget for revisions, not a rev A.
  • The government track has an 18–30 month lead time. Start the credential work now so it matures alongside the technology, not after it.

8. Sequencing with gates

GateProof requiredUnlocks
G1 — Legibilityrfd live with 10 RFDs; spirit-abi v1 published; conformance suite runs in CIExternal review, citation, contribution
G2 — Accountabilityoracle replays a Spirit + agent session bit-exact; attest produces a verifiable receiptResearch publication; procurement conversations
G3 — IndependenceThird party builds a Spirit from published docs and passes conformanceClaim “platform” honestly; Imaginarium launch
G4 — SovereigntyNode R0 boots measured into VUDO; two nodes converge over LoRa after an IP partitionOpen-hardware release; mission pitch with a demo
G5 — MissionOne civic service running on the stack with published controlsContract track, grant renewal, the political argument made materially

Nothing after G3 is worth attempting before G1 is done. G1 costs weeks and is the highest-leverage work in this document.


Sources for the comparative claims: Oxide’s public repositories and RFD archive; samhsmith.com and serenum.computer release notes; Handmade Seattle 2023 demo. Univrs state as recorded in project context — re-verify before relying on specifics.

Sources

  • Univrs project context and internal architecture notes (VUDO / DOL / Skills, mycelial-network crates, ENR economics) — re-verify specifics against the live repos before relying on them.
  • Handmade Seattle 2023 demo (referenced for the systems-programming / hand-built-tooling lineage).
  • Comparative platform reads used for Matrices A–C draw on the design philosophies and published limitations of the incumbent agent/cloud platforms discussed in-text.

Part of the Univrs research ecosystem: