Reasoning fragments
Candidates can conflict, correlate, or converge without independent evidence.
Story 01 · The premise
Govern intelligence. Prove consequence.
Coordinate reasoning. Govern authority. Verify action. Reconcile outcomes.
Govern privatelyExecute locallyProve selectively
IntentAuthorityActionReceipt
Stories
Four ways into the control plane.
Follow the narrative, or jump directly to the operating evidence.
The shift
AI is becoming an organization, not a single assistant.
01Model
→
02Assistant
→
03Agent
→
04Agent team
→
05Organization
The new system
- Specialized agents, models, and compute strategies
- Parallel candidate generation and execution
- Shared memory, tools, and destinations
- Internal and external counterparties
As intelligence becomes heterogeneous and parallel, the scarce layer is the control plane that decides what runs, under whose authority, and what evidence counts.
The control-plane problem
More agents create more coordination risk.
Authority blurs
Identity, role, scope, policy, and budget are not bound to each action.
Execution disappears
Tool calls and off-platform outcomes are not read back independently.
Evidence breaks
Approvals, artifacts, decisions, and results cannot be reconciled into one trace.
Boundaries fail
A counterparty cannot verify a shared action without receiving the private record.
Parallel intelligence without governed routing and verification produces speed without accountability.
Wrk.Flo architecture
A control plane above models, agents, tools, and workflows.
modelsagentscomputetoolsworkflowssystems
Aggregate + reconciliation spineresults merged · contradictions resolved · governed state advanced
Eden OS
The intended human interface for building and overseeing the control plane.
Learn
Understand the organization, goals, roles, systems, and operating context.
Build
Compose agent teams, workflows, policies, integrations, and evidence contracts.
Operationalize
Assign recurring work, memory, schedules, dependencies, and service levels.
Execute
Route and run approved work across models, tools, and external systems.
Oversee
Monitor, approve, intervene, audit, correct, and improve.
Eden OS · human-facing organizational interfaceWrk.Flo · governed control plane beneath
System report
The detailed operating model.
Architecture, authority, evidence, implementation status, and the limits of the proposed verification rail.
Repository reality vs. Apex build
Separate implementation signals from planned integrations.
- Tenant-scoped runtime and trace identifiers
- Approval bindings to normalized targets and action hashes
- Destination readback and correction / recovery patterns
- Local proof-of-agency prototype and receipt contracts
- Bounded parallel candidates with evidence-aware consensus
- Bounded Proof-of-Agency credential
- Public-key verifier and selective disclosure
- Cardano preprod receipt commitment with rollback handling
- One buyer-validated cross-organization workflow
- Security, privacy, legal, and economic review
Repository evidence is a starting substrate, not proof of production Cardano integration or external security review.
Governed orchestration loop
Every action moves through a governed loop.
IntentDecompose into bounded lanes
Lane ARead / query
Lane BExternal write
Lane CIrreversible
target statecapability + authoritypolicy / decision versionrun / evidence receiptcancellation state
Destination readback
→Independent verification
→Reconcile
→Learn
Authority to act and authority to declare success are separate.
Agent-to-agent trust
Machine-to-machine work needs an authority handshake.
Authority envelope · request
- Issuer and subject
- Agent or actor identity
- Requested capability
- Policy version and limits
- Approval requirements
- Evidence and receipt expectations
- Validity, expiry, revocation, correlation IDs
Response envelope
- Accepted, denied, or escalated
- Result and destination reference
- Evidence hashes
- Correction or rollback reference
- Signed receipt
Communication is not enough. Each machine action needs inspectable authority, policy, evidence, and state.
Verifiable action mechanism
Every consequential action should produce a reconcilable receipt.
- who
- organization + accountable human + acting agent
- authority
- credential + role + policy version
- decision
- exact action + approval state
- evidence
- input / output roots + destination readback
- state
- completed · blocked · reversed · superseded
Original
Correction
Revocation / reversal
Current governed view
Receipts hold commitments and references, not raw prompts, customer data, identity documents, credentials, or private artifacts.
Hybrid trust boundary
Centralized authority. Selective verification.
Inside the control plane
- Identity, tenant data, memory, artifacts
- Policy, approvals, budgets, execution authority
- Credential issuance, recovery, correction
- Authoritative private state
Minimized commitments
- Credential status commitments
- Selected receipt-batch Merkle roots
- Expiry, revocation, correction, supersession refs
- Independent signature and inclusion checks
Cardano rail
- Candidate independent timestamp and integrity layer
- Preprod anchoring only during prototype
- No execution power, customer data, or diagnosis
- Optional when simpler proof is sufficient
Private authority
Portable proof
No relying party should have to trust the operator’s mutable database, but Cardano remains optional unless it solves that problem better than signed logs or an independent timestamp service.
Interactive ecosystem map
See where authority, execution, and evidence meet.
Select a layer to inspect its role, control boundary, and evidence output.
Control plane
Wrk.Flo coordinates the system without moving private authority onto a public rail.
Current implementation + Apex target
Wrk.Flo
control plane
control plane
Beachhead hypothesis
Make cross-organization evidence handoffs verifiable.
Primary buyerRegulated service provider
Economic buyerHead of Compliance or Enterprise Risk
Human release approval→Bounded agent action→Destination readback→Private evidence package→Signed receipt→Cardano commitment
Apex validation targetTest whether one buyer-confirmed workflow can reduce evidence-handoff burden without expanding agent authority.
Evidence and commercial path
Evidence today. Commercial proof next.
Repository-level workflow builder, authority checks, approval bindings, destination readback, recovery patterns, and local receipt prototypes.
No production Cardano integration, external security review, named customer validation, or audited commercial evidence is claimed here.
Pilot implementation
Map one consequential workflow, integrate systems and evidence, define authority and recovery.
Recurring platform
Operationalize orchestration, governance, memory, and workflow operation through Eden OS.
Enterprise verification
Credential lifecycle, selective disclosure, and cross-organization receipt services where buyer validation justifies them.
The near-term wedge is the mapping from private authority, policy, evidence, destination readback, correction, and recovery into portable verification, not a generic token or financial product.
Leadership statement
“Delegate more work to AI without giving up human accountability.”
Moses Tut · CEO & Co-Founder
Team
Founder-market fit across AI, institutions & Web3.
Moses Tut
CEO & Co-Founder
- Former Managing Director, gener8tor OnRamp
- Former Draper University program ops & fundraising
- Former Pact Consortium partnerships — blockchain-enabled lending
- Operator across workforce access, venture support & governed AI
Isaac Tut
CTO & Co-Founder
- AI product & systems engineering
- Boston Scientific experience
- Technical leadership across multi-agent infrastructure
community & workforce systemsaccelerator & venture opsWeb3 partnershipsgoverned AI infrastructure
Selected prior team experience — background only; not an endorsement, investment, or partnership.
The accountability stack
Three pillars. One governed system.
Governed multi-agent orchestration
Eden OS is the intended human interface for classifying actions and binding consequential work to an accountable organization, human, role, policy, and exact decision.
classify→bind→approve→execute→read back→recover
Tamper-evident selective verification
For selected consequential actions and RWA-grade provenance needs: signed receipts, minimized commitments, Merkle batching, and a candidate Cardano proof rail.
Not every action is on-chain. Blockchain does not establish identity, legal authority, or outcome truth.
Quantum optimization
A forward-looking optimization layer for routing, scheduling, resource allocation, and candidate search across complex multi-agent workloads.
Research direction only. Not a current MVP capability or accelerator claim.govern actionprove consequenceoptimize next
The synthesis
Govern intelligence. Prove consequence. Optimize what comes next.
Eden OS governs multi-agent action. Selected receipts can cross trust boundaries. Quantum optimization remains the horizon, not the MVP.
Private where authority lives. Tamper-evident where proof must travel. Human accountability at the center.
01 · GovernOrganization + accountable human + exact decision
02 · VerifySelected signed receipts + portable provenance
03 · OptimizeFuture routing + scheduling + resource allocation