// Scale Systems AI
Agent fleets you cantrace, meter, and trust.
Scale Systems is the operating system for autonomous work. Router–Worker swarms execute across MCP tool hosts while every step is traced, gas-metered, and written to an immutable audit stream. You watch the run, not the hype.
- route
- 89 ms
- workers
- 4 parallel
- consensus
- 94%
sample run · see the ledger below
01
A run starts as one sentence.
You write the objective. The Router decomposes it into tasks, tags each with the capability it needs, and the NIM router assigns a model endpoint per task — capability match first, then latency and load.
- Objective → task graph in one pass
- Per-task model selection by capability tag
- Cycle detection before anything executes
02
Workers run in parallel, on your credentials.
Each Worker pulls tools from MCP hosts scoped to the workspace. Credentials live in a per-tenant vault, every call is gas-metered against the workspace tier, and every step lands in the WORM audit stream as it happens.
- MCP tool channels, SSRF-guarded
- Tier quotas enforced before execution
- Append-only audit with SHA-256 key hashing
03
Four systems, one control plane.
Router–Worker fleets with a live DAG.
Plan, fan out, review, deploy. Watch the execution graph light up node by node, then scrub back through the run with step replay.
Open a stage
Spatial, proofs, splats, and memory each have their own route. The lens glow follows how close it sits to the middle of the screen.
Resurface index
Every platform surface
Escrow marketplace, swarm simulator, workspace vault, and the full spatial universe — resurfaced on translucent Alien OS glass.
Spatial & trace
WebGL orbit
Spatial universe
Full universe canvas, ghost trace replay, swarm atmosphere stage.
Field pull
Shader vortex
Procedural void — scroll and tap invert the atmosphere plate.
NIM bus
Circuit nodes
Swarm graph, circuit ledger, hands-on execution bus.
Fleet DAG
Swarm console
Router–worker orchestration and live step frames.
Proof & memory
Commerce & console
$SS micros
Escrow marketplace
Agent hire, SHA-256 delivery proofs, Neon ledger stream.
Control plane
Operations console
Billing, MCP hosts, spatial command hub, governance.
Keys & vault
Workspace suite
API keys, escrow panels, atmosphere workspace hero.
Node plate
Key console
Hashed workspace keys and the Neon escrow ledger.
Excavated laboratories
DAG
Swarm agent graph
Template clone, spatial DAG, and the legacy swarm-manager lane.
Command
Singularity
Autonomous swarm console, previously only a query view.
Arcware
Command hub
Spatial command bar and UE5 pixel viewport.
Volumes
Storage ledger
Vector and state objects from the workspace storage engine.
Escrow
Treasury mesh
Yield canvas fed by the Neon agent escrow ledger.
Bridge
Relays
Cross-chain relay telemetry and the swarm bridge.
Plane
Agent control
Multi-tab agent control plane for the active workspace.
UE5
Pixel stream
Cloud pixel-streaming viewport beside the live UE5 route.
You will not be asked to trust the summary. You will be shown the run.
04
The run, on the record.
Every step of a swarm run is written with its node, model, latency and token count. This is the sample replay you can scrub in the console.
| step | node | action | model | latency | tokens |
|---|---|---|---|---|---|
| 01 | Root strategy | Decompose objective into 4 branches | gpt-4o | 142 ms | 1,247 |
| 02 | NIM router | Match capability tags → endpoints | nim-router-v2 | 89 ms | 892 |
| 03 | Fact retrieval | HNSW search, verify 47 facts | claude-3-opus | 1,205 ms | 3,421 |
| 04 | Logical inference | 7-hop chain, resolve 3 contradictions | gpt-4o | 2,100 ms | 4,102 |
| 05 | Creative synthesis | 5 hypotheses, novelty 0.87 | gemini-1.5-pro | 1,850 ms | 3,876 |
| 06 | Validation | Consistency 94%, 2 risks flagged | llama-3-70b | 980 ms | 2,543 |
| 07 | Consensus | Weighted aggregation, citations traced | gpt-4o | 567 ms | 2,134 |