CONCEPT

JSON-OS

programs → protocols → intelligence

Thesis

Programs become protocols. The OS becomes an AI agent that plans, negotiates, and executes contracts against tools. AGI or not, the UX feels post-app: users state goals; the agent composes tool contracts, executes them in sandboxes, and writes receipts to a ledger.

Kernel_Primitives

Contract

JSON Schema that declares a tool's intents, inputs/outputs, costs, side-effects.

Capability

Attenuated, signed grants (ocap-style) the agent presents to tools.

Planner

Converts user goals → contract graph (DAG) with budgets & policies.

Scheduler/Runner

Executes nodes in WASM/web-container sandboxes.

Ledger

Append-only event log (local-first, CRDT sync) with receipts/attestations.

Policy Engine

JSON policies capping cost, data egress, PII, regions, retries.

Tool Bus

Adapters for MCP servers (e.g., Coronium MCP Proxy Agent), HTTP, DBs.

Identity

Task-scoped keys/DIDs; ephemeral by default.

Payments

HTTP-402 / agent-commerce adapters for per-use resources.

json_os_flow.sh

$ user: "analyze sales data and send report"

→ planner: construct_contract_graph()

├─ node_1: fetch_data(db.sales)

├─ node_2: analyze(model=gpt-4)

└─ node_3: send_email(to=team@co.com)

→ policy_check: cost_ok, pii_allowed

→ scheduler: parallel_execute(sandboxed)

✓ completed: ledger_receipt_#x7k2m

Explore JSON-OS→