Eight layers, one environment you own.
Efficient Labs isn't one feature — it's the Ownership Layer for autonomous intelligence: eight foundational layers, from the models you bring to the money your agents may spend, each labelled with what it can actually prove today. The file system is the source of truth; everything else is a projection of it.
- Layers
- Eight
- Source of truth
- Your file system
- Proof
- Signed receipts
- Status
- Honestly labelled
From the model you bring to the money it may spend
Read it as a stack. Intelligence is any model you bring; Identity establishes who owns and who acts; Continuity (ECP)preserves it across time; StratosAgentruns it locally; the Governance Harnessbounds every action; Atmosphere makes it yours to own; ARI measures readiness; and Commerce governs spend. Each is labelled with the maturity it has actually reached — never rounded up.
Claude, GPT, Gemini, Qwen, DeepSeek, open-weights — frontier or local. Models are commodities; you bring them, swap them, and never get locked in.
See the docsWho owns this intelligence, which node and which agent is acting, what authority and permissions exist. Identity is the first requirement for accountability.
See the docsContext, state, knowledge, receipts, ledgers and manifests that persist — turning temporary AI outputs into durable, compounding organizational intelligence.
See the docsRun autonomous intelligence locally, under your control — model routing, tool and workflow execution, identity, receipts and governance. Open source, free forever.
Get StratosAgentCapability, authority and human-approval gates, escalation and evaluation policies, receipts, audit trails and denial records — turning autonomous into trustworthy autonomous.
See the docsOwn your compute, context, memory, identity and infrastructure — intelligence that lives outside cloud dependency. Sovereignty instead of data sharecropping.
Explore AtmosphereMeasure whether intelligence is ready for autonomy — across governance, ownership, continuity, security, economics, commerce, identity and infrastructure — and what it costs.
Run your readiness indexThe money-governance layer for autonomous work: should this transaction occur, who authorized it, what budget and wallet govern it, what limits and evidence apply — governed spend before autonomy ever touches money.
See the docsWhat actually ships against the layers
The layers are the frame; these are the products that implement them. Each is owned by you, open where it counts, and labelled with its real maturity. DOP — the Digital Organism Protocol — is the improvement metabolism that decides what proves out and what gets retired.
StratosAgent
Intelligence Infrastructure
Run autonomous intelligence locally — model routing, tools, workflows, identity, receipts, governance. Free forever.
Learn moreECP
Continuity Infrastructure
Preserve and compound intelligence — manifests, ledgers, receipts, read models. Outputs become durable, reusable intelligence.
Learn moreGovernance Harness
Trust Infrastructure
Govern and prove every action — authority, approvals, protected-action gates, deny-by-default, audited denials.
Learn moreAtmosphere
Ownership Infrastructure
Replace cloud dependency with sovereign infrastructure — own your compute, context, memory and intelligence.
Learn moreARI
Measurement Infrastructure
Measure whether intelligence is ready for autonomy across eight dimensions — and what it costs.
Learn moreDOP
Improvement Metabolism
Records outcomes, evaluates work, promotes what proves value, prunes what decays — humans keep control of evolution.
Learn moreCommerce
Financial Infrastructure
Govern budgets, wallets, approvals and settlement — financial accountability before autonomy touches money.
Learn moreThree guarantees that compose
The whole architecture rests on three properties working together. Each is useful alone; the combination is what lets nodes that don't know each other cooperate safely. Trust the properties, not the participants.
Content-addressed
- SHA-256 freshness model
- Data named by its contents
- Tampering changes the name
Capability-isolated
- Broker env-scoping strips secrets
- WASI sandbox, empty preopens
- Write-approval 402 loop
Post-quantum sealed
- ML-DSA-65 signatures
- On skills and receipts
- Forward-safe provenance
Why all three, together
Content-addressing proves what a payload is. Post-quantum seals prove who vouched for it. Capability isolation contains what it can do once it runs. With all three, a node can accept code and data from a peer it has never met — because it never has to trust the peer, only the math.
How a skill crosses the mesh safely
A skill is the unit of capability that moves between nodes. Its lifecycle is built so that no node ever has to trust where a skill came from — only that its seal verifies and its execution is contained.
1 · Sealed at origin
2 · Gossiped peer-to-peer
3 · Verified on ingest
4 · Run sandboxed
Self-evolution stays inside the box
Self-authored skills are confined to a narrow integer-transform class — sealed and observed like any other. The agent can extend itself, but never outside that class.
How we label every capability
Full status matrix →- Live
- Running in production, exercised by tests.
- Wired
- Built and connected into the daemon; hardening in progress.
- Standalone
- Built and tested in isolation; live wiring is supervised.
- Mock
- Scaffold / placeholder — explicitly NOT real yet.
Automate the architecture, not the wrapper
The design principle under all eight layers: a correctly-designed file and dataflow architecture does the work deterministically, cheaply, and auditably. The agent earns its place only where genuine ambiguity lives — not as a wrapper around everything.
Deterministic by default
Inference where it counts
Auditable end to end
Go deeper
An architecture you can verify.
See the mesh that carries it, the agent that runs on top, or the full honest status of every capability.