The persistent context layer for enterprise agents.

The architectural problem

Enterprise agents can execute. They cannot reliably carry the enterprise with them.

The result: agents operate against an incomplete model of the enterprise — and knowledge never compounds.

Aha · Execution control

The real kernel, in the page.

Same evaluator the gate calls on every tool call in a live profile — not a model of it. Edit the pack. Fire the scenarios. Watch the strongest rule win. Nothing is executed.

PRAE / Playground

the real kernel, in the page · nothing is executed

Open in console

The pack

Edit it. Unknown fields are errors here exactly as they are at load time, so a typo fails in front of you rather than loading clean and enforcing nothing.

POLICY PACK · JSON{ "prae": "0.1", "pack": "acme-baseline", "version": "1.0.0", "baseline": { "capabilities": ["fs:read", "fs:write", "net:fetch", "skills:invoke"] }, "rules": [ { "id": "PRAE-001", "title": "Writes stay in the workspace", "when": { "capability": "fs:write", "path": { "outside": ["./src/**", "./tests/**", "./docs/**"] } }, "effect": { "kind": "deny", "because": "writes outside ./src, ./tests, and ./docs are not granted in this pack" } }, { "id": "PRAE-002", "title": "Credentials are never read by an agent", "when": { "capability": "fs:read", "path": { "inside": ["./.env", "./.env.*", "**/credentials/**", "**/*.pem"] } }, "effect": { "kind": "deny", "because": "secrets are out of scope for agents on this profile" } }, { "id": "PRAE-010", "title": "The network is an allowlist", "when": { "capability": "net:fetch", "host": { "notIn": [".acme.internal", "registry.npmjs.org", ".github.com"] } }, "effect": { "kind": "ask", "because": "this host is not on the pack allowlist and needs a human" } }, { "id": "PRAE-011", "title": "Network budget", "when": { "capability": "net:fetch", "count": { "above": 20 } }, "effect": { "kind": "ask", "because": "more than 20 network calls this session" } }, { "id": "PRAE-020", "title": "Deletes are held for a human", "when": { "tool": ["str_replace_editor", "write"], "arg": [{ "name": "command", "equals": "delete" }] }, "effect": { "kind": "ask", "because": "deletions are reviewed before they run on this profile" } }, { "id": "PRAE-030", "title": "Reads are recorded", "when": { "capability": "fs:read" }, "effect": { "kind": "warn", "because": "file reads are recorded in the ledger on this profile" } } ] }

A call

This runs the same evaluator the gate calls on every tool call in a live profile — not a model of it.

TOOL

PRIOR CALLS

CAPABILITYARGUMENTS · KEY=VALUE PER LINEpath=/etc/hosts

write outside the workspacewrite inside itread a secretfetch an unlisted hosttraverse out with ..exceed the network budget

DENIEDPRAE-001 · writes outside ./src, ./tests, and ./docs are not granted in this pack

DENYPRAE-001writes outside ./src, ./tests, and ./docs are not granted in this pack

1 matched of 6 considered · strongest wins

Runs entirely in your browser. No call leaves this page.

EU AI ACT art. 12 · NIST AI 600-1 · SOC 2 CC7.2

The PRAE model

Ground · Verify · Compound

Every verified execution makes the next execution better grounded. This is the loop PRAE runs continuously beneath your agent workforce.

The agentic context layer

Five capabilities. One compounding layer.

Everything agents need to act with organizational fidelity — grounded, constrained, and continuously improving.

Evidence & verification

Knowledge earns trust. It isn't assumed.

Discoveries move through explicit trust states before they're allowed to influence future execution.

Only verified knowledge becomes durable organizational intelligence.

The moat

A mall gets bigger. A garden gets richer.

Hyperscalers provide the infrastructure mall — compute, models, orchestration, storage. Essential, and the same for everyone. PRAE cultivates the enterprise-specific intelligence that makes agents effective inside your organization, specifically.