v0.1 Engineering Blueprint

ARM Agents

The Agentic AI Operating System. A futureproof, failproof durable execution fabric for the Agentic Internet Era era.

Protocol-native MCP / A2A / ACP / AP2 with zero-loss resilience, carbon-aware orchestration, and human-aligned safety by design.

99.99%
Recovery SLA
MCP/A2A/AP2
Protocols
Wasm
Sandbox
3-Model BFT
Consensus
Scroll to explore
01

Vision & North Star

ARISE exists to make the 2026 Agent Internet era safe, sustainable, and abundant for everyone.

MISSION // Build the foundational execution fabric that enables billions of AI agents to collaborate safely across organizational boundaries, creating a Silicon Workforce that amplifies human potential while respecting planetary boundaries. Every transaction auditable, every mandate revocable, every watt accounted for.

Silicon Workforce

Autonomous agent swarms handle routine work, freeing humans for creative, strategic, and relational endeavors.

Human-in-the-Loop Safety

Cryptographically signed mandates with revocable sub-delegation ensure humans always retain ultimate authority.

Planetary Sustainability

Carbon-aware scheduling routes computation to green-grid windows. Every agent action carries an emissions trace.

Equitable Access

Open protocols and anti-extractive safeguards prevent platform lock-in. Agent capabilities as public infrastructure.

02

Core Engineering Principles

Ten invariants that govern every design decision in the ARISE codebase.

P1

Sovereign Agency

Agents act within cryptographically bounded mandate scopes. No action without verifiable authorization.

P2

Protocol-Native

MCP, A2A, ACP, and AP2 are first-class citizens, not adapters. Wire formats are the source of truth.

P3

Durable State

Every state transition is checkpointed. Zero-loss recovery from crashes, network partitions, or provider failures.

P4

Carbon-Aware

Computation is scheduled against grid carbon intensity. Non-urgent tasks defer to green windows automatically.

P5

Auditability by Design

Immutable Truth Ledger records every decision, tool call, and mandate chain for EU AI Act compliance.

P6

Semantic Sandboxing

Each tool executes in a capability-restricted Wasm sandbox. Tools cannot escape their declared permission scope.

P7

Economic Traceability

AP2 mandates create verifiable payment trails. Every transaction is attributable to a human principal.

P8

Graceful Degradation

When AI reasoning fails, the system falls back to deterministic paths rather than hallucinating forward.

P9

Composable Swarms

Agents discover and coordinate via A2A/ACP without central registries. Swarm topology is emergent.

P10

Edge-Ready

Core kernel compiles to Wasm for edge/IoT deployment. Full system runs on 512MB RAM minimum.

03

High-Level Architecture

Layered execution fabric with orthogonal cross-cutting concerns for economics, sustainability, and auditability.

L0

Human Intent Layer

DID IdentityRoot MandateHITL EscalationRevocable Sub-Mandates
L1

Orchestration Kernel

Temporal WorkflowsTask RouterDead-Man's SwitchCircuit Breakers
L2

Safety & Sandbox Layer

Wasm SandboxesBFT ConsensusHallucination DetectionMandate Verifier
L3

Cognitive Swarms (A2A/ACP)

Agent DiscoveryTask LifecyclePerformativesStructured Dialogue
L4

Memory Fabric

Vector StoreEvent SourcingCheckpoint StoreSemantic Cache
L5

MCP Tools & Actions

Tool RegistryContext WindowsResource ProvidersPrompt Templates
CROSS-CUTTING

Economic Layer (AP2)

Mandate Signing
Payment Rails
Tx Ledger

Sustainability Scheduler

Grid Intensity API
Carbon Budgets
Green Windows

Truth Ledger

Immutable Log
Decision Traces
Audit Queries

DATA FLOW

Human Intent (top-down)
Tool Results (bottom-up)
Economic signals (lateral)
Audit traces (all layers)
ARCHITECTURE GRAPH (Mermaid TD)
graph TD
    HI[Human Intent / DID] --> OK[Orchestration Kernel]
    OK --> SS[Safety & Sandbox Layer]
    SS --> CS[Cognitive Swarms - A2A/ACP]
    CS --> MF[Memory Fabric]
    MF --> MT[MCP Tools & Actions]
    MT --> ACT[Real-World Actions]

    OK -.-> EL[Economic Layer - AP2]
    OK -.-> SUS[Sustainability Scheduler]
    OK -.-> TL[Truth Ledger]

    EL -.-> MT
    SUS -.-> OK
    TL -.-> SS

    style HI fill:#0d9488,stroke:#0d9488,color:#000
    style OK fill:#1a1f2e,stroke:#0d9488
    style SS fill:#1a1f2e,stroke:#334155
    style CS fill:#1a1f2e,stroke:#334155
    style MF fill:#1a1f2e,stroke:#334155
    style MT fill:#1a1f2e,stroke:#334155
04

Key Capabilities

Protocol-mapped capabilities with concrete implementation strategies and measurable success criteria.

CapabilityProtocol / TechImplementation NotesImpact / Metric
Durable Workflow ExecutionTemporal.io / Custom Rust RuntimeEvent-sourced state machines with checkpoint-at-every-step. Replay from any point after crash.99.99% task recovery rate
MCP Tool IntegrationMCP (Anthropic Spec)Native MCP server/client. Tools register with JSON Schema. Context windows managed per-agent.<50ms tool dispatch latency
Agent Discovery & CoordinationA2A (Google DeepMind).well-known/agent.json auto-discovery. AgentCard + task lifecycle (submitted/working/done/failed).<200ms cross-org agent handshake
Structured DialogueACP (Agent Communication Protocol)FIPA-style performatives (request, propose, inform, refuse). Dialogue state tracked per conversation.100% dialogue state consistency
Economic TransactionsAP2 (Agent Payment Protocol)Verifiable mandate chains. Payment-agnostic (crypto/fiat/credits). Extends A2A task artifacts.Full economic traceability
BFT Plan Consensus3-Model AgreementHigh-stakes plans require 2-of-3 model agreement. Disagreements trigger HITL escalation.>95% false positive rejection
Wasm Tool SandboxingWasmtime / WasmEdgeEach tool runs in capability-restricted Wasm module. Memory limits, syscall filtering, time bounds.Zero sandbox escapes
Carbon-Aware SchedulingWattTime / Electricity Maps APINon-urgent tasks defer to low-carbon grid windows. Real-time gCO2/kWh tracking per task.30% compute carbon reduction
Truth Ledger AuditingImmutable Append-Only LogEvery decision, tool call, mandate invocation logged with Merkle proofs. EU AI Act compliant.Full audit trail, <1s query
Hallucination DetectionCross-Reference + GroundingOutput grounding against tool results. Confidence scoring with automatic fallback to deterministic paths.<2% undetected hallucinations
05

Resilience & Failproof Mechanisms

Six interlocking patterns that make zero-loss execution possible under Byzantine conditions.

Stateful Task Objects

A2A Task lifecycle (submitted -> working -> input-required -> done | failed) stored durably. Any crash replays from last checkpoint, not from scratch.

// Temporal workflow — checkpoint on every state transition
#[workflow]
async fn agent_task(ctx: WorkflowCtx, task: A2ATask) -> Result<TaskResult> {
    let plan = ctx.activity(plan_action, &task).await?;     // checkpoint 1
    let approved = ctx.activity(bft_consensus, &plan).await?; // checkpoint 2
    
    if !approved.quorum_met {
        ctx.activity(escalate_hitl, &plan).await?;           // checkpoint 3
        return Ok(TaskResult::InputRequired);
    }
    
    let result = ctx.activity(execute_sandboxed, &plan).await?; // checkpoint 4
    ctx.activity(log_truth_ledger, &result).await?;             // checkpoint 5
    Ok(TaskResult::Done(result))
}

BFT 3-Model Consensus

High-stakes decisions require agreement from 3 independent models. Byzantine Fault Tolerant: tolerates 1 compromised/hallucinating model. Disagreement triggers human escalation.

// BFT consensus for high-stakes plan approval
async fn bft_consensus(plan: &AgentPlan) -> ConsensusResult {
    let models = [Model::Primary, Model::Secondary, Model::Auditor];
    let votes: Vec<Vote> = join_all(
        models.iter().map(|m| evaluate_plan(m, plan))
    ).await;
    
    let approvals = votes.iter().filter(|v| v.approved).count();
    ConsensusResult {
        quorum_met: approvals >= 2,  // 2-of-3 agreement
        votes,
        confidence: votes.iter().map(|v| v.confidence).sum::<f64>() / 3.0,
    }
}

Dead-Man's Switch

Detects agent loops, spirals, and runaway execution. Monitors token burn rate, action repetition, and wall-clock time. Triggers graceful shutdown + HITL alert.

// Dead-Man's Switch — kills runaway agents
struct DeadManSwitch {
    max_steps: u32,
    max_tokens: u64,
    max_wall_time: Duration,
    repetition_window: u32,
}

impl DeadManSwitch {
    fn check(&self, ctx: &AgentContext) -> SwitchResult {
        if ctx.steps > self.max_steps { return SwitchResult::Kill("Step limit") }
        if ctx.tokens_burned > self.max_tokens { return SwitchResult::Kill("Token limit") }
        if ctx.elapsed() > self.max_wall_time { return SwitchResult::Kill("Timeout") }
        if ctx.action_entropy() < 0.1 { return SwitchResult::Kill("Loop detected") }
        SwitchResult::Continue
    }
}

Semantic Wasm Sandboxing

Tools execute in Wasm modules with capability-restricted interfaces. Declared permissions (network, filesystem, memory) enforced at runtime. No ambient authority.

// Wasm sandbox configuration per tool
{
  "tool": "web_search",
  "sandbox": {
    "memory_limit_mb": 64,
    "cpu_time_limit_ms": 5000,
    "permissions": {
      "network": ["GET https://*.googleapis.com/*"],
      "filesystem": "none",
      "env_vars": "none"
    },
    "syscalls": ["clock_gettime", "random_get"],
    "output_schema": { "type": "array", "items": "SearchResult" }
  }
}

Circuit Breakers

Provider failures trigger circuit breakers with exponential backoff. Automatic failover to backup providers. Half-open probes test recovery before full re-enable.

// Circuit breaker state machine
enum CircuitState { Closed, Open(Instant), HalfOpen }

impl CircuitBreaker {
    async fn call<F, T>(&mut self, f: F) -> Result<T>
    where F: Future<Output = Result<T>> {
        match self.state {
            CircuitState::Open(until) if Instant::now() < until => {
                Err(Error::CircuitOpen)
            }
            CircuitState::Open(_) => {
                self.state = CircuitState::HalfOpen;
                self.try_call(f).await
            }
            _ => self.try_call(f).await,
        }
    }
}

Graceful Degradation

When AI reasoning confidence drops below threshold, the system switches to deterministic fallback paths. No hallucinating forward — safe defaults always available.

// Graceful degradation with confidence thresholds
async fn execute_with_fallback(task: &Task) -> Result<Output> {
    let ai_result = ai_engine.reason(task).await;
    
    match ai_result.confidence {
        c if c > 0.85 => Ok(ai_result.output),
        c if c > 0.60 => {
            log_truth_ledger("low_confidence", &ai_result);
            escalate_hitl(task, &ai_result).await
        }
        _ => {
            log_truth_ledger("fallback_triggered", &ai_result);
            deterministic_fallback(task).await  // Safe, tested path
        }
    }
}
06

Future-Proofing & Interoperability

Abstraction layers and extension points designed for a decade of protocol evolution.

Reasoning Engine Abstraction

The cognitive layer is fully decoupled from the execution kernel via a CognitiveProvider trait. Swap GPT-5 for Claude Opus or a local Llama model with zero workflow changes.

trait CognitiveProvider: Send + Sync {
    async fn reason(&self, ctx: &ReasoningContext) -> ReasoningResult;
    async fn evaluate_plan(&self, plan: &Plan) -> PlanEvaluation;
    fn capabilities(&self) -> ProviderCapabilities;
}

// Hot-swap at runtime
kernel.set_cognitive_provider(Arc::new(ClaudeProvider::new()));

Agent Self-Discovery

Every ARISE agent publishes a .well-known/agent.json conforming to the A2A AgentCard spec. Includes capabilities, supported protocols, auth requirements, and rate limits.

// .well-known/agent.json
{
  "name": "supply-chain-optimizer",
  "version": "1.2.0",
  "protocols": ["mcp/1.0", "a2a/1.0", "acp/1.0", "ap2/1.0"],
  "capabilities": ["planning", "negotiation", "payment"],
  "auth": { "type": "oauth2", "issuer": "https://auth.arise.dev" },
  "rate_limit": { "rpm": 1000, "burst": 50 },
  "carbon_budget": { "gco2_per_task": 2.5 }
}

Protocol Upgrade Paths

Version negotiation built into every protocol handshake. Agents advertise supported versions and negotiate to highest common. Breaking changes use adapter layers.

// Version negotiation during A2A handshake
async fn negotiate_protocol(peer: &AgentCard) -> ProtocolVersion {
    let my_versions = vec!["a2a/1.1", "a2a/1.0"];
    let peer_versions = &peer.protocols;
    
    my_versions.iter()
        .find(|v| peer_versions.contains(v))
        .map(|v| ProtocolVersion::parse(v))
        .unwrap_or(ProtocolVersion::BASELINE)
}

Open Extension Points

Plugin system for custom tools, cognitive providers, storage backends, and protocol adapters. Extensions are Wasm modules loaded at runtime with sandboxed capabilities.

// Plugin manifest for a custom tool extension
{
  "plugin": "salesforce-connector",
  "type": "mcp-tool",
  "wasm_path": "./plugins/salesforce.wasm",
  "permissions": ["network:salesforce.com", "memory:128mb"],
  "schema": "./plugins/salesforce-schema.json",
  "hooks": ["pre_execute", "post_execute", "on_error"]
}
07

Security, Compliance & Ethical Layer

Threat model, regulatory compliance, and ethical safeguards for enterprise and governmental deployment.

Threat Model

ThreatAttack VectorMitigationSeverity
Agent SpoofingForged AgentCard or DID impersonationMutual TLS + DID verification on every A2A handshake. AgentCards signed by issuing authority.Critical
Mandate CoercionCompromised agent signs sub-mandates beyond scopeMandate chain verification at every execution boundary. Scope intersection enforced cryptographically.Critical
Sandbox EscapeWasm tool exploits host function interfaceMinimal host function surface. Capability-based access. Memory isolation. Syscall allowlists.High
Hallucination PropagationHallucinated output used as input to downstream agentBFT consensus on critical paths. Grounding verification. Confidence-gated execution.High
Economic DrainAgent authorized for payments exploits mandate boundsAP2 mandate amount caps. Rate limiting. Real-time spend monitoring. Sub-second revocation.High
Data ExfiltrationTool sends context window data to unauthorized endpointNetwork allowlists per tool sandbox. Egress monitoring. Content classification scanning.Medium

Regulatory Compliance

EU AI Act (2026)

High-risk AI system registration and documentation
Human oversight mechanisms (HITL escalation)
Transparency: users informed when interacting with AI
Fundamental rights impact assessments
IMPL Truth Ledger provides full audit trail. Mandate chains document human oversight. Agent disclosure headers on all A2A messages.

Carbon / ESG Reporting

Per-task energy consumption tracking
Scope 2 emissions attribution
Carbon budget enforcement
Third-party verifiable sustainability claims
IMPL Kepler + WattTime integration. Per-agent carbon accounting in Truth Ledger. Exportable ESG reports in GRI/SASB formats.
08

Tech Stack & Integration Patterns

Production-grade toolchain optimized for safety, performance, and protocol compliance.

Core Runtime

Rust

Kernel, orchestrator, protocol implementations

Mojo

High-performance inference pipelines, numeric workloads

Wasmtime

Tool sandboxing, edge deployment, plugin runtime

Orchestration

Temporal.io

Durable workflow execution, checkpointing, replay

NATS

Inter-agent messaging, event streaming

gRPC/tonic

High-performance RPC for internal service mesh

Storage & Memory

SurrealDB

Multi-model store (graph + document + KV) for agent state

Pinecone

Vector embeddings for semantic memory, RAG retrieval

FoundationDB

Ordered KV for Truth Ledger, Merkle-based audit log

Protocol Clients

mcp-rs

Rust MCP client/server (Anthropic spec)

a2a-sdk

A2A task lifecycle, agent discovery, AgentCard

acp-core

ACP performatives, dialogue management

Observability

OpenTelemetry

Distributed tracing, metrics, spans across agent swarms

Prometheus

Metrics collection, alerting, SLA monitoring

Grafana

Dashboards, Truth Ledger visualization, carbon tracking

Sustainability

WattTime API

Real-time grid carbon intensity data

Electricity Maps

Fallback carbon intensity provider, regional data

Kepler (k8s)

Per-pod energy consumption measurement

Integration Code Examples

MCP Server Registrationrust
use mcp_rs::{McpServer, ToolDefinition, JsonSchema};

#[tokio::main]
async fn main() {
    let server = McpServer::builder()
        .name("arise-tools")
        .version("1.0.0")
        .tool(ToolDefinition {
            name: "web_search".into(),
            description: "Search the web with sandboxed execution".into(),
            input_schema: JsonSchema::from_file("schemas/web_search.json"),
            handler: Box::new(|input| Box::pin(web_search_handler(input))),
        })
        .tool(ToolDefinition {
            name: "database_query".into(),
            description: "Execute read-only SQL against approved schemas".into(),
            input_schema: JsonSchema::from_file("schemas/db_query.json"),
            handler: Box::new(|input| Box::pin(db_query_handler(input))),
        })
        .build();

    server.listen("0.0.0.0:3001").await.unwrap();
}
A2A Task Submissionrust
use a2a_sdk::{A2AClient, Task, TaskState, AgentCard};

async fn submit_cross_org_task(peer_url: &str) -> Result<Task> {
    let client = A2AClient::new();
    
    // Discover agent capabilities
    let card: AgentCard = client
        .discover(peer_url)  // GET /.well-known/agent.json
        .await?;
    
    // Submit task with mandate chain
    let task = client.submit_task(
        &card,
        Task::builder()
            .title("Optimize Q3 supply chain routing")
            .mandate(current_mandate_chain())
            .input(serde_json::json!({
                "constraints": ["cost < $50k", "delivery < 72h"],
                "regions": ["NA", "EU"]
            }))
            .build()
    ).await?;

    // Poll for completion (or use streaming)
    assert_eq!(task.state, TaskState::Submitted);
    Ok(task)
}
AP2 Mandate Flowrust
use ap2::{Mandate, SubMandate, PaymentIntent, SigningKey};

fn create_payment_mandate() -> Mandate {
    let root_key = SigningKey::from_env("ROOT_MANDATE_KEY");
    
    let root = Mandate::root()
        .principal("did:key:z6Mk...")  // Human DID
        .max_amount(Currency::USD(10_000))
        .expires_at(Utc::now() + Duration::hours(24))
        .sign(&root_key);

    // Delegate sub-mandate to procurement agent
    let sub = SubMandate::delegate(&root)
        .agent("did:agent:procurement-bot-7")
        .max_amount(Currency::USD(5_000))
        .allowed_actions(["purchase_order", "negotiate_price"])
        .revocable(true)
        .sign(&root_key);

    // Agent creates payment intent within mandate bounds
    let payment = PaymentIntent::new()
        .mandate(&sub)
        .amount(Currency::USD(2_350))
        .recipient("did:vendor:acme-supplies")
        .memo("PO-2026-Q3-0042")
        .build();
    
    root
}
12

Interactive Truth Ledger Explorer

A live demo of the immutable audit log. Every agent decision, tool call, and payment is recorded, verifiable, and queryable.

6
Entries
6/6
Verified
0.274 gCO2
Total Carbon
100%
Chain Integrity
END-TO-END WORKFLOW TRACE
StrategicPlannerVendorScoutBFT-ValidatorComplianceGuardPaymentExecutorStrategicPlanner

Complete audit trail for mandate procurement-q1 showing 6 recorded steps across 4 agents with full Merkle chain verification and HITL escalation at payment threshold.

15

Real-World Workflow: Cross-Company Procurement

An end-to-end example showing how two companies negotiate a logistics contract using A2A discovery, ACP negotiation, BFT validation, HITL approval, and AP2 payment.

Acme Corp
ARISE Registry
GlobalLog Inc.
ARISE Safety

Workflow Summary

Protocols Used
5 (A2A, ACP, AP2, BFT, HITL)
Organizations
3 (Acme, GlobalLog, ARISE)
Agent Handoffs
10 steps, fully traced
Safety Checks
BFT + HITL + Mandate

Every step is recorded in the Truth Ledger with Merkle proofs. The human stays in control via mandate chains and HITL escalation. The payment only executes after BFT consensus, human approval, and cryptographic mandate verification. Click any step above to see the protocol-level detail.

14

Why ARISE?

Feature-by-feature comparison with the leading agent frameworks. ARISE is the only protocol-native, resilient, carbon-aware agent OS.

FeatureARISELangGraphCrewAIAutoGenSemantic Kernel
Protocol Support
Native MCP support
Google A2A protocol
Agent Commerce (AP2)
FIPA-ACL (ACP)
Resilience
Durable execution (stateful)
Dead-Man's Switch
Multi-model BFT consensus
Circuit breakers
Safety & Isolation
Wasm sandboxing
Mandate chain (delegated auth)
Immutable audit log
HITL escalation
Sustainability
Carbon-aware scheduling
Per-task energy tracking
EU AI Act compliance
Architecture
Edge deployment (512MB)Q3
Multi-org agent discovery
Plugin marketplaceQ3
Full support
Partial / community plugin
Q3Planned
Not available

The Core Differentiators

Protocol-Native

ARISE is the only framework with first-class MCP + A2A + ACP + AP2 support, enabling true cross-organization agent collaboration and commerce.

Zero-Loss Resilience

Durable execution via Temporal, BFT consensus, Dead-Man's Switch, and circuit breakers. No other framework offers this depth of fault tolerance.

Carbon-Aware by Design

Built-in carbon scheduling, per-task energy tracking, and ESG reporting. Sustainability is architecture, not an afterthought.

09

Implementation Roadmap

Four-phase delivery plan from MVP kernel to ecosystem marketplace.

Phase 1Active

Resilient Kernel + MCP

Months 0-6

MILESTONES

  • Temporal-based durable execution kernel in Rust
  • Native MCP server/client with tool registry
  • Wasm sandbox runtime (Wasmtime integration)
  • Dead-Man's Switch + circuit breaker patterns
  • Truth Ledger v1 (append-only FoundationDB)
  • OpenTelemetry instrumentation baseline

RISKS

  • Temporal Rust SDK maturity
  • Wasm sandbox escape vectors

SUCCESS METRICS

  • 99.99% task recovery rate
  • < 50ms tool dispatch
  • Zero sandbox escapes
Phase 2

A2A/ACP + AP2 Protocols

Months 6-12

MILESTONES

  • A2A agent discovery (.well-known/agent.json)
  • Task lifecycle management (full A2A spec)
  • ACP performative engine (request/propose/inform)
  • AP2 mandate signing + payment intents
  • BFT 3-model consensus module
  • Cross-org task integration tests

RISKS

  • A2A spec stability (pre-1.0)
  • Payment rail regulatory variance

SUCCESS METRICS

  • < 200ms cross-org handshake
  • First production AP2 transaction
  • 100% mandate traceability
Phase 3

Carbon Scheduler + Edge

Months 12-18

MILESTONES

  • WattTime/Electricity Maps integration
  • Carbon-budget-aware task scheduler
  • Wasm kernel compilation for edge devices
  • Edge-cloud hybrid orchestration
  • Kepler per-pod energy measurement
  • EU AI Act audit report generation

RISKS

  • Carbon API data quality/latency
  • Edge memory constraints

SUCCESS METRICS

  • 30% compute carbon reduction
  • 512MB minimum edge footprint
  • Full EU AI Act compliance
Phase 4

Ecosystem & Marketplace

Months 18-24

MILESTONES

  • Plugin marketplace for Wasm extensions
  • Agent reputation and trust scoring
  • Multi-tenant SaaS deployment option
  • Robotics/IoT agent bridge protocols
  • Community-contributed tool registry
  • Open-source core release (Apache 2.0)

RISKS

  • Marketplace governance
  • Multi-tenant isolation guarantees

SUCCESS METRICS

  • 100+ community plugins
  • 10+ enterprise deployments
  • 1000+ registered agents
10

Launch Platform

The user-facing control plane for operating, auditing, and monitoring the ARISE agent fleet.

Protocol Status Dashboard

Real-time status of all protocol endpoints (MCP, A2A, ACP, AP2) with health checks, latency percentiles, and connection counts.

WebSocket live updates
5-minute rolling P99 latency
Active agent count
Protocol version distribution

Truth Ledger Explorer

Browse, search, and verify the immutable audit log. Every decision trace, tool invocation, and mandate chain is queryable.

Merkle proof verification
Full-text search on decision logs
Mandate chain visualization
Export for EU AI Act audits

Agent Auditing Console

Inspect any agent's execution history, mandate scope, resource consumption, and compliance status.

Per-agent activity timeline
Mandate scope tree view
Resource consumption graphs
Compliance score card

Carbon Dashboard

Track energy consumption and carbon emissions across the agent fleet. Budget enforcement and scheduling optimization visibility.

Real-time gCO2/kWh by region
Per-task carbon attribution
Budget vs actual tracking
Green window utilization %

Fleet Observability

OpenTelemetry-powered distributed tracing across agent swarms. Traces span organizational boundaries via A2A correlation.

Cross-org trace stitching
Flame graphs per workflow
Error rate dashboards
SLA compliance monitoring

Agent Discovery Registry

Browse all discoverable agents, their capabilities, protocol support, and trust scores. The Yellow Pages of the Agent Internet.

Capability search
Trust score rankings
Protocol compatibility matrix
Rate limit visibility

LIVE PROTOCOL STATUS

MCP v1.0Operational
23ms
P99 latency
847
Active agents
A2A v1.0Operational
89ms
P99 latency
312
Active agents
ACP v0.9Beta
45ms
P99 latency
156
Active agents
AP2 v0.8Alpha
112ms
P99 latency
42
Active agents
13

Business & Ecosystem Model

Multi-layered revenue strategy designed to fund open-source development while serving enterprises, governments, and the developer community.

ARISE Cloud (Managed SaaS)

Fully managed multi-tenant deployment with per-agent and per-workflow pricing. Ideal for teams that want to run agents without managing infrastructure.

PRICING Free tier: 5 agents, 1K tasks/mo. Pro: $0.002/task + $5/agent/mo. Enterprise: Custom.
TARGET Startups, SMBs, dev teams

Enterprise & Sovereign License

Self-hosted or air-gapped deployment for organizations with strict data residency, regulatory, or security requirements. Includes priority support and SLAs.

PRICING Annual license: $50K-$500K based on agent fleet size. Includes 24/7 support, compliance toolkit, and custom integrations.
TARGET Governments, defense, finance, healthcare

Efficiency Tax (AP2 Transactions)

A micro-fee on economic transactions processed through the AP2 payment protocol. Revenue scales directly with agent-driven commerce volume.

PRICING 0.1% on AP2 transaction value (capped at $10/tx). Volume discounts for >$1M/mo.
TARGET Cross-org agent commerce

Governance-as-a-Service

Managed compliance reporting, EU AI Act audit generation, Truth Ledger hosting, and regulatory certification support for organizations operating agent fleets.

PRICING Starting at $2K/mo for audit log hosting. Enterprise bundles include penetration testing and compliance certification.
TARGET Regulated industries

Plugin Marketplace (Rev Share)

A curated marketplace for Wasm-sandboxed agent extensions, tool adapters, and pre-built workflows. Developers earn 80% revenue share on paid plugins.

PRICING 20% platform commission on paid plugins. Free listing for open-source plugins.
TARGET Developer ecosystem, ISVs

Training & Certification

Structured training programs, certification tracks, and consulting engagements for teams adopting ARISE. Includes architecture reviews and deployment planning.

PRICING Online courses: $500-$2K. On-site workshops: $15K-$50K. Architecture review: custom.
TARGET Enterprise teams, system integrators

TARGET UNIT ECONOMICS (YEAR 2)

CAC (Cloud)
$120
Developer-led growth, docs-first GTM
LTV (Pro)
$8,400
28-month avg retention, 15% expansion
LTV:CAC
70:1
Target >10:1 at scale
Gross Margin (Cloud)
78%
Rust runtime efficiency + Wasm sandboxing
Gross Margin (License)
92%
Marginal cost is support staff
NRR Target
135%
Agent fleet growth drives natural expansion

Open-Source Core Commitment

The ARISE kernel, protocol adapters (MCP/A2A/ACP/AP2), Wasm sandbox runtime, and Truth Ledger core are permanently open-source under Apache 2.0. Enterprise features (multi-tenancy, SSO, advanced compliance tooling, priority support) are source-available under a commercial license.

This model ensures the protocol layer remains a public good while sustaining development through value-added services. We believe the agent internet must be built on open standards, not proprietary lock-in.

Apache 2.0
Core runtime
Source-Available
Enterprise features
80/20 Rev Share
Plugin marketplace
Free Tier
Always available
11

Risks, Mitigations & Testing Strategy

Comprehensive risk register with concrete mitigation strategies and testing approaches.

Protocol specification instability

P: HighI: Medium

MITIGATION

Version negotiation layer. Adapter pattern for breaking changes. Active spec working group participation.

TESTING STRATEGY

Protocol conformance test suite against multiple spec versions. Fuzz testing on wire format parsing.

Wasm sandbox vulnerabilities

P: MediumI: Critical

MITIGATION

Minimal host function surface. Defense-in-depth: Wasm + process isolation + seccomp. Regular security audits.

TESTING STRATEGY

Continuous fuzzing with cargo-fuzz. Annual third-party penetration testing. Bug bounty program.

Temporal.io single point of failure

P: LowI: High

MITIGATION

Multi-region Temporal deployment. Custom lightweight replay engine as fallback. Regular DR drills.

TESTING STRATEGY

Chaos engineering: Temporal cluster kill tests. Failover time measurement. Data integrity verification after recovery.

Carbon API data quality

P: MediumI: Low

MITIGATION

Dual provider (WattTime + Electricity Maps). Fallback to conservative estimates. Cache with TTL.

TESTING STRATEGY

Historical data backtesting. API response time monitoring. Comparison audits between providers.

Multi-model BFT false positives

P: MediumI: Medium

MITIGATION

Tunable consensus thresholds. Human review queue for edge cases. Continuous calibration from HITL feedback.

TESTING STRATEGY

Adversarial test suite with known-hallucination prompts. False positive rate tracking dashboard.

Supply chain attacks on plugins

P: MediumI: High

MITIGATION

Plugin signing + verified publisher chain. Static analysis on Wasm bytecode. Runtime behavior monitoring.

TESTING STRATEGY

Malicious plugin test suite. Provenance chain verification. Sandbox escape attempt library.

16

Get Started

From zero to a running agent fleet in under 5 minutes. Apache 2.0 licensed, open-source core.

1

Install the CLI

# Install ARISE CLI
curl -fsSL https://get.arise.dev | sh

# Or via npm
npm install -g @arise/cli
2

Initialize your first node

# Create a new ARISE project
arise init my-agent-fleet

cd my-agent-fleet

# This scaffolds:
#   /agents       - Agent definitions
#   /tools        - MCP tool configs
#   /mandates     - Mandate templates
#   /workflows    - Temporal workflow defs
#   arise.config.ts
3

Define and register an agent

// agents/procurement.ts
import { defineAgent } from "@arise/sdk"

export default defineAgent({
  name: "ProcurementAgent",
  protocols: ["mcp", "a2a", "acp"],
  capabilities: ["vendor_search", "price_compare"],
  mandateScope: "procurement/*",
  consensus: { model: "bft", quorum: "2/3" },
  carbon: { budget: "5g_co2_per_task" },
})
4

Start your local agent fleet

# Start in development mode
arise dev

# Output:
# ARISE v0.1.0 - Development Mode
# Kernel:    http://localhost:9100
# Dashboard: http://localhost:9200
# MCP:       http://localhost:9300
# A2A:       http://localhost:9400
#
# Agent 'ProcurementAgent' registered
# Truth Ledger: append-only (local SQLite)
# Carbon tracking: mock provider (dev mode)