Platform architecture
See how workspace scope, evidence, encryption, and durable agent work fit together.
Sebastian's finance AI security architecture uses tenant data isolation by workspace schema, applies row-level security to user-private records, and uses accounting data encryption for selected sensitive fields. Human-in-the-loop accounting controls can pause designated writes before execution.
Workspace business data lives in a tenant-specific Postgres schema selected by the authenticated request context.
workspace scopedGlobal identity and workspace membership are resolved before services receive their scoped database handle.
request boundaryRow-level security protects user-private tables; workspace-shared finance records remain available to workspace members by design.
explicit scopeAgent messages, checkpoints, memory, filenames, and connector credentials use per-workspace AES-256-GCM keys.
field inventoryDesignated write tools can pause execution before the tool runs. The journal tool creates a draft in pending review.
human checkpointCanonical SHA-256 digests allow stored close source snapshots to be re-verified on the server.
content checkSebastian does not present repository controls as SOC 2, ISO, or auditor assurance without documentary proof.
proof requiredPer-workspace field encryption protects a defined sensitive-field inventory, not every ordinary finance column.
scoped encryptionServer-enforced preparer-versus-reviewer identity checks remain an implementation gate.
not yet claimedA durable accounting-period lock and documented reopen workflow are not represented as shipped controls.
implementation gateSee how workspace scope, evidence, encryption, and durable agent work fit together.
Inspect source provenance, control totals, content hashing, and server verification.
Evaluate security alongside provenance, authority, exceptions, and monitoring evidence.
AI for accounting and FP&A on the ledger you already run. No migration. No rip and replace.