
The orches|
One CLI to create, supervise, and scale AI agents. Built in Rust. WASM-sandboxed. Erlang-style fault tolerance. Token-cost optimization that saves you money on every call.
From lifecycle management to cost optimization, Nexus OS provides the primitives that production agent systems require.
One-for-one, one-for-all, rest-for-one restart strategies. Agents self-heal with configurable restart windows and escalation policies.
Multi-step orchestration with automatic compensation. If step 3 fails, steps 1 and 2 are automatically rolled back.
Per-agent budgets, automatic model downsizing, semantic caching, and prompt compression. Every call is routed to the cheapest capable handler.
Deterministic tasks run in WASM sandboxes at near-zero cost. Skills are portable, versioned, and composable.
Every agent action is logged with Lamport timestamps, content hashes, and cost tracking. Full replay and fork capabilities.
Deploy agents to Cloudflare Workers for sub-50ms latency worldwide. Offline-first with automatic sync.
Conflict-free replicated data types for agent memory. Merge without coordination. Fork and replay from any point.
Skill → WASM → LLM routing cascade. The broker picks the cheapest handler that can solve the task with high confidence.
Nexus OS organizes agent systems into three distinct layers. The Execution Layer runs code in WASM sandboxes. The Orchestration Layer manages supervisors, sagas, and workflows. The Intelligence Layer handles LLM routing and cost optimization.


The naos CLI is your single interface for the entire agent lifecycle. Initialize projects, create agents from templates, deploy with supervisors, track costs, and monitor everything from your terminal.
naos initScaffold a new project in < 5 minutesnaos createCreate agents from built-in templatesnaos run / stopManage agent lifecyclenaos deployDeploy with supervisor strategiesnaos dashboardLaunch the web dashboardnaos cost statusView real-time cost trackingnaos broker routeRoute tasks through the brokernaos edge deployDeploy agents to Cloudflare Edgenaos audit tailStream the causal audit trailA built-in web dashboard served by the naos binary itself — no separate install needed. Server-rendered HTML with auto-refresh, terminal-aesthetic styling, and 5 JSON API endpoints.

/Real-time summary of agents, supervisors, cost, trust, and recent activity
/agentsAll agents with status, cost budget, and AXIS trust scores
/supervisorsSupervisor trees with child agent states and restart counts
/sagasMulti-step saga definitions with execution history and rollback status
/workflowsSequential data pipeline steps with execution tracking
/poolsConcurrent fan-out pools with member status and strategy
/costBudget tracking with progress bars and per-agent spend breakdown
/auditSearchable, filterable log of every system event
/trustTrust verification status for all registered agents
/brokerRouting stats, skill calls, WASM/LLM breakdown, and cost savings
/edgeCloudflare Worker deployments, global latency, region distribution
/api/agents/api/supervisors/api/cost/api/audit/api/trust/api/broker/api/edgeThe broker evaluates every task against registered skills, WASM modules, and LLM providers. It picks the fastest, cheapest handler that meets the confidence threshold — saving up to 90% on token costs.
broker routebroker executebroker statsbroker configbroker skillsOne command pushes your agent to Cloudflare Workers with Durable Objects for persistent state. Sub-50ms latency worldwide, automatic scaling, and zero cold starts.
edge loginedge deployedge listedge statusedge logsedge undeployDefine your entire agent system in a single nexus.config.yaml. Agents, supervisors, cost budgets, broker routing, skills, and deployment targets — all in one place.
The config file supports agent templates (research, coding, data, custom), supervisor strategies with restart windows, per-agent cost budgets with automatic throttling, broker routing with skill/WASM/LLM cascade, and edge deployment targets.
project: my-agents
version: "1.0"
agents:
researcher:
template: research
model: claude-sonnet
capabilities:
- resource: "web.*"
actions: [read]
- resource: "fs.output"
actions: [write]
supervisor:
strategy: one-for-one
max_restarts: 5
window: 300s
cost:
researcher:
budget: $10/day
alert_at: 80%
action: throttle
broker:
routing:
preferSkill: true
llmAsLastResort: true
thresholds:
skill: 0.9
wasm: 0.8
skills:
- name: summarize
patterns: [summarize, summary, tldr]
handler: "fn:summarize_text"
cost: "$0.0001"
edge:
cloudflare:
account_id: "${CF_ACCOUNT_ID}"
zone_id: "your-zone-id"
routes:
researcher: "researcher.agents.nexus/*"Nexus OS is MIT licensed. The CLI, dashboard, and all orchestration primitives are free. You only pay for the LLM tokens your agents consume.
Install the CLI, initialize a project, create your first agent, and watch it run.
Nexus OS supports macOS and Linux on x86_64 and ARM architectures.
Install directly from GitHub with Cargo. Requires Rust toolchain.
Download a pre-built binary from GitHub Releases for your platform.
Clone the repository and build with Cargo for full control.
Verify your installation: naos --version