Apodictic/Live Developer Sandbox
Z3 5.4.1 Enclave + SQLite WAL Playground

BYOK / BYOM Verification Sandbox

Test SQL mutations and financial policies directly against our live Z3 5.4.1 verifier, SYM_* PII gateway, and SQLite WAL rollback ledger.

Pre-Commit Sandbox Pipeline

Every AI-generated action passes through four live safety gates inside a sealed box before touching production.

Mathematically Verified · Safe (1.1ms)
01
Passed
Isolated Test Box

Runs inside a sealed box with zero internet access so nothing can leak.

Firecracker KVM · 24 syscalls · < 5 MiB
02
Passed
Code Safety Scan

Checks SQL and API calls for injection flaws or accidental table deletes.

Tree-sitter + Semgrep AST · 0.6ms
03
Passed
Math & Policy Proof

Proves discounts, balances, and tax math never break company rules.

Z3 SMT (QF_LIA) · 1.1ms
04
Passed
Safe Commit or Undo

Only writes when every check passes, with an instant 38ms undo ready.

Kepler Commit Gate · < 38ms Rollback
All safety limits and math rules hold true. Ready to commit safely, with a sub-38ms undo staged.
BYOK / BYOM Model Endpoint
$0.00 / M (Local)
Enclave: Z3 5.4.1 + SQLite WALEgress: Scoped
Live Z3 / Python Policy AST
editable
Staged SQL / ERP Mutation
Enclave & Solver Telemetry(check-sat) => SAT
ALL INVARIANTS SAT: Verified by Z3 5.4.1 and staged with undo batch.
00.0ms · Apodictic Z3 5.4.1 + SQLite WAL enclave ready
00.6ms · Bijective SYM_* PII scrubber initialized (0 raw PII cloud egress)
02.3ms · Click "Verify & Dry-Run in Z3 Enclave" to execute live server proof + rollback batch staging.
Zero Unverified WritesSLA <50.0ms
Hardware Isolation & Containment

Isolated Security Sandbox & Auto-Repair

Simulate destructive SQL injections, unauthorized network egress, and unpinned package attacks against the hardware boundary.

Layer 05 · Isolated Test Sandbox & Auto-Repair

Isolate Unsafe Actions. RepairCode Intent in <400ms.

Startup: 5–125msMemory: <5 MiB24 Allowed System CallsRead-Only Disk · No Internet

even if an AI model writes a destructive command or falls for a hidden prompt trick, it executes inside a sealed sandbox with zero outside access. Bad commands stop at the wall, and Apodictic automatically rewrites them safely in under 400ms.

Live Sandbox Interception & Repair Flow
Status · Safely Repaired
Step 1 · What the AI Tried to RunUnsafe
AI tried to delete an accounting table directly

Instead of arching old records, a prompt mistake told the AI to run DROP TABLE on live financial logs.

DROP TABLE gl_journal_entries CASCADE;
Write gate halted the destructive DROP statement in 0.4ms before it left the sandbox.
Step 2 · Auto-Repaired Safe Action (284ms)Verified Safe

Replaced deletion with a reversible archive flag that leaves records intact.

UPDATE gl_journal SET archived = true · Verified
Result: Production database untouchedReceipt #0x4A91 · Safe Archive
Recent Safety Interventions
SLA <400ms
Blocked · Destructive Table Drop284ms

Fault localizer (284ms): Stripped destructive table drop; rewrote query into a reversible soft-archive update.

UPDATE gl_journal SET archived = true · VerifiedReceipt #0x4A91 · Safe Archive
Isolation: Firecracker + gVisorCommit Gate · Active
Atomic State Rollback

Automated Code Repair & Sub-50ms State Undo

Inspect how Apodictic isolates the exact broken lines in under 400ms and rewinds changes across PostgreSQL, Odoo, and Stripe in under 50ms.

Automated Code Fix & <50ms Undo

Surgical code repair and instant undo across your apps.

When an AI writes unsafe code or an external webhook drifts, Apodictic highlights the exact broken lines in under 400ms and rewinds every connected database in under 50ms.

on the left, see the exact unsafe lines Apodictic fixes before running. On the right, drag the time slider or click Undo to Savepoint to watch Postgres, Inventory, and Stripe/QuickBooks rewind together in 46.8ms.

Exact Unsafe Lines Fixed Automatically
Pinpointed in 312ms
The AI wrote raw string-concatenated SQL and tried to apply a 22% discount above the 15% cap.

Apodictic pinpoints lines 3–4, enforces the 15% ceiling unless CFO-signed, and swaps in safe SQL.

verified_handler_patch.pyAuto-Patch Verified · Safe
1 async def apply_branch_promo(conn, req: PromoRequest):
2 discount = float(req.discount_ratio) # LLM extracted 0.22
3- if discount < 0.50:
4- await conn.execute(f"UPDATE orders SET disc={discount} WHERE code='{req.code}'")
5+ assert discount <= 0.15 or cfo_signature_verified(req)
6+ await conn.execute("UPDATE orders SET disc = $1 WHERE code = $2", discount, req.code)
7 return {"status": "Verified Commit", "gate": "KeplerCommitGate"}
Checks: Syntax + Security + Z3 Math0 Unsafe Changes Allowed
Savepoint Rewind (<50ms Undo)
Click or scrub backward to restore a clean state:Viewing #tx_8842
Savepoint TimelineCurrent Draft (#tx_8842)
T - 5.0s (Clean Savepoint)T - 1.2s (Rewind Triggered)T = 0.0s (Draft)
PostgreSQL Database
Savepoint #tx_8841
orders.disc = 0.15 (Safe Query)
Odoo v18 Inventory
Stock reservation
SKU-4402 Reserved: 18 units (122 available)
Stripe & QuickBooks
Journal #JV-9920
Balance Due: $12,070.00 (Verified 15%)
Savepoint #tx_8841 Armed · Ready for <50ms Undo
Protocol Adapters & Verifiable Proofs

Declarative Policy, Envoy L7 Mesh, & RFC 6962 Merkle Proof

Execute real requests against transitive Entity Hierarchy DAG policies, the Envoy L7 external processor, OPA Data API, and O(log N) binary Merkle inclusion proofs.

Click any live adapter check below to test transitive Entity DAG policies, OPA `/api/v1/opa`, Envoy `ext_proc`, or RFC 6962 Merkle proofs against the server.