HARCH|CORP
HarchOS /0.1
HarchOS is the operating system for Africa's first carbon-aware GPU compute platform — powered by Morocco's renewable energy advantage. Open-source from day one.
Planned across 5 hubs
Morocco-wide mesh
Target carbon intensity
Why HarchOS /0.2
The AI infrastructure gap in Africa isn't just about hardware — it's about sovereignty, sustainability, and scale.
AI workloads in Africa route through US and EU data centers by default. Every inference call, every training run, every dataset leaves the continent. HarchOS keeps data on African soil — by default, not by configuration. Every deployment enforces data residency at the infrastructure level with three sovereignty tiers: strict, regional, and global.
Training a single LLM can emit over 300 tonnes of CO2. On a fossil-heavy US grid, that number multiplies. Morocco's grid runs on solar, wind, and CSP — averaging 47 gCO2/kWh versus 400+ in the US. HarchOS doesn't just run workloads — it schedules them when and where carbon intensity is lowest, without sacrificing performance.
Africa has 1% of global compute capacity for 17% of the population. That gap is widening, not closing. HarchOS is building the infrastructure to reverse this trend — starting from Morocco's strategic position at the crossroads of Europe and Africa, with undersea cables reaching both continents.
Hub Network /0.3
Each hub is strategically located near renewable energy sources. Combined, they form Africa's largest planned GPU mesh — 1,798 GPUs across solar, wind, and hybrid installations.
Solar CSP — Noor Complex
GPUs
800
Renewable
97%
Offshore Wind — Atlantic Coast
GPUs
400
Renewable
85%
Solar + Wind Hybrid — Green City
GPUs
350
Renewable
75%
Wind + Grid — Mediterranean Gateway
GPUs
200
Renewable
65%
Grid + Solar — Urban Edge
GPUs
48
Renewable
40%
Hub specifications are planned targets. Actual deployment timelines depend on infrastructure development and regulatory approval.
Architecture /0.4
HarchOS operates on a three-layer architecture inspired by autonomous systems — observe the grid, predict the optimal path, execute with precision. Every layer is open-source and auditable.
Continuous ingestion of carbon intensity data from national grid APIs, energy spot pricing from Moroccan utilities, GPU utilization metrics from all hubs, and inter-hub network reliability telemetry. The SENSE layer maintains a real-time topology graph of every GPU, including NVLink bandwidth, PCIe lane assignments, and cooling capacity.
The ML-powered scheduling engine that predicts the optimal hub, time window, and GPU configuration for each workload. It considers carbon intensity forecasts, spot pricing trends, workload latency requirements, and sovereignty constraints simultaneously — optimizing for the Pareto frontier of cost, carbon, and performance.
Deploys, migrates, and scales workloads automatically based on THINK layer decisions. Supports live migration between hubs when carbon conditions shift — moving a training job from Tanger to Ouarzazate when solar output peaks. Zero-touch operations with self-healing and automated failover.
Open Source /0.5
The entire HarchOS platform is open-source. Server, SDKs, CLI, Terraform provider, Grafana plugins, examples — everything. We believe sovereign infrastructure requires transparent code.
Every component is developed in the open on GitHub under permissive licenses (Apache 2.0, MIT, MPL). You can audit, fork, contribute, or self-host. Sovereign infrastructure can't be sovereign if you can't read the code that runs it.
9
Public repositories
5
Languages (Python, TS, Go)
3
Package registries (PyPI, npm, TF)
7
Open-source repos
Python SDK /0.6
The official Python SDK provides OpenAI-compatible inference with built-in carbon tracking. Every API response includes carbon footprint data — gCO2 emitted, renewable percentage, and the hub that served your request. Supports async workloads, SSE streaming, and a context manager for carbon-aware batch processing.
pip install harchosv0.2.1 on PyPIfrom harchos import HarchOS
client = HarchOS(api_key="hsk_...")
# OpenAI-compatible inference
response = client.chat.completions.create(
model="harchos/h100-llama3",
messages=[{"role": "user", "content": "Hello"}],
region="morocco",
sovereignty="strict"
)
# Built-in carbon tracking
print(response.carbon_footprint.gco2) # e.g. 2.4
print(response.carbon_footprint.renewable) # e.g. 0.97
print(response.carbon_footprint.hub) # e.g. "ouarzazate"TypeScript SDK /0.7
The TypeScript/JavaScript SDK brings sovereign defaults to Node.js. Every client is initialized with region='morocco', sovereignty='strict', and carbonAware=true — you have to explicitly opt out of carbon-awareness, not opt in. Full type safety, carbon intensity queries, optimal hub finder, and workload scheduling.
npm install @harchos/sdkimport { HarchOSClient } from "@harchos/sdk";
const client = new HarchOSClient({
apiKey: "hsk_...",
region: "morocco",
sovereignty: "strict",
carbonAware: true // default: true
});
// Carbon intensity query
const intensity = await client.carbon
.getIntensity("MA");
console.log(intensity.gco2PerKwh); // ~47
// Find optimal hub
const hub = await client.carbon
.optimalHub({ gpus: 8, type: "H100" });
// Deploy workload
const workload = await client.workloads
.create({ image: "pytorch/pytorch", gpus: 4 });CLI /0.8
Written in Go for single-binary cross-platform distribution. Sovereign by default, multi-profile configuration, structured output (JSON/YAML/table), and CI/CD-friendly with --dry-run and --quiet modes. Secure credential storage via system keyring. Available via Homebrew, binary download, or build from source.
brew install HarchCorp/tap/harchos# Authenticate
$ harchos auth login
✓ Authenticated as user@harchcorp.com
# Deploy a workload
$ harchos workloads deploy \
--image pytorch/pytorch \
--gpus 8 --type H100 \
--region morocco \
--sovereignty strict
✓ Workload deployed to ouarzazate
ID: wrk-7f3a2b1c
Hub: ouarzazate (97% renewable)
Carbon: ~2.1 gCO2/kWh
# Scale up
$ harchos workloads scale wrk-7f3a2b1c --gpus 16
# Migrate to greener hub
$ harchos workloads migrate wrk-7f3a2b1c --hub dakhlaTerraform /0.9
Manage HarchOS workloads, models, inference endpoints, datasets, network policies, and storage volumes with Terraform. Define sovereignty levels in HCL — and once set, sovereignty cannot be downgraded. The provider enforces this at the API level. Available on the Terraform Registry as HarchCorp/harchos.
terraform init && terraform applyterraform {
required_providers {
harchos = {
source = "HarchCorp/harchos"
version = "~> 0.1"
}
}
}
# Deploy with strict sovereignty
resource "harchos_workload" "training" {
name = "llm-finetune"
image = "pytorch/pytorch:latest"
gpus = 8
gpu_type = "H100"
region = "morocco"
sovereignty = "strict" # cannot be downgraded
carbon_aware = true
max_gco2 = 50 # gCO2/kWh threshold
}
data "harchos_hubs" "available" {
region = "morocco"
min_renewable = 75
}Grafana /0.10
Six custom Grafana plugins for GPU monitoring, carbon tracking, hub health, workload distribution, and carbon forecasting. Introduces two custom query languages: EnergyQL for energy metrics and SovereigntyQL for compliance auditing.
Connect Grafana directly to HarchOS hubs. Query real-time GPU utilization, carbon metrics, hub health, and sovereignty status with custom query languages.
Visualize GPU utilization across all hubs. Drill down by hub, rack, or individual GPU. Track memory, compute, and thermal metrics in real time.
Track carbon emissions per workload, per hub, or per project. Compare against grid averages and renewable targets. Historical trend analysis.
Monitor hub operational status including power supply, cooling capacity, network connectivity, and GPU availability across the mesh.
Visualize how workloads are distributed across hubs. See carbon-aware routing decisions in real time. Identify optimization opportunities.
Predict future carbon intensity based on weather forecasts, solar/wind generation predictions, and grid demand patterns. Plan workloads accordingly.
Carbon Intelligence /0.11
HarchOS doesn't just run workloads — it understands the carbon impact of every compute decision and actively minimizes it.
Every API response includes a carbon_footprint object with gCO2 emitted, renewable percentage, and the serving hub. Track your carbon impact per request, per session, or per project with zero additional API calls.
ML models predict carbon intensity 24-48 hours ahead using weather forecasts, solar generation curves, wind speed predictions, and historical grid patterns. Schedule training jobs when carbon intensity will be lowest.
The THINK layer automatically routes workloads to the greenest available hub. If Ouarzazate's solar output peaks at 2PM, your training job gets routed there. If Dakhla's wind picks up at midnight, batch processing shifts automatically.
Generate carbon reports per project, per team, or per organization. Export in formats compatible with GHG Protocol, CDP, and EU CSRD reporting frameworks. Built-in audit trail for regulatory compliance.
US Fossil Grid
400+
gCO2/kWh average
EU Average Grid
250
gCO2/kWh average
HarchOS Mesh
~47
gCO2/kWh target
Sovereignty /0.12
Data sovereignty isn't a feature — it's the architecture. HarchOS enforces sovereignty at three levels, and once set, it can never be downgraded.
Data and workloads never leave the designated hub. All processing, storage, and inference happen within a single facility. Maximum data isolation for government, defense, and regulated industries.
Workloads can move between hubs within the same geographic region (Morocco). Enables carbon-aware routing within the country while keeping data on sovereign soil. Recommended for most enterprise workloads.
Workloads can span multiple regions across Africa. Enables pan-African mesh networking for distributed training and low-latency inference across the continent. Data still never leaves Africa.
Sovereignty levels are enforced at the infrastructure level and cannot be downgraded once set. A workload deployed with strict sovereignty will never be migrated to a regional or global configuration — even under failover conditions. This is a technical guarantee, not a policy.
Specifications /0.13
These are our engineering targets, not current capabilities. We believe in transparency over marketing. Every number here is a commitment we're working to deliver.
Security /0.14
Every request is authenticated and authorized — no implicit trust between services, hubs, or users. Mutual TLS between all hub components. Service mesh with per-request identity verification.
Workloads and data never leave the designated region. Technical enforcement at the infrastructure level, not just policy. Encryption keys are region-bound and cannot be exported. Verified by automated compliance checks.
TLS 1.3 in transit, AES-256 at rest. HSM-backed key management in every hub. Keys never leave the sovereign region. Support for customer-managed keys (BYOK) for maximum control.
Designed for GDPR, Morocco's Law 09-08, and SOC 2 Type II from the ground up. Compliance isn't bolted on — it's built into the architecture. Automated audit trails and continuous compliance monitoring.
Roadmap /0.15
Our roadmap is public because accountability matters. Every phase has clear deliverables and measurable outcomes.
2025
2026
2027
2028+
GitHub /0.16
Every component of HarchOS is developed in the open. Audit the code, file issues, contribute patches, or deploy your own instance.
Join the waitlist for HarchOS early access. Deploy on Africa's first carbon-aware GPU cloud — or explore the open-source codebase today.