Rillet

The systems the agents act across are named: a Native Integrations product line sits in the platform navigation as the single source of financial data, the site footer discloses Rillet as an agent of Plaid Financial Ltd for regulated bank account information, the info hub names Salesforce, Stripe, Ramp, Brex, Rippling and Snowflake as the connected stack with continuous bank and processor syncing and native Stripe payout matching, and the Aura page has workflow agents handling the data pulls and the MCP surface reaching Salesforce and Snowflake.

SourceRillet, rillet.com/product/aura-ai, rillet.com/product/native-integrations named in navigation, and info.rillet.comread 2026-09-07

Customers compose multi step workflows that agents run: the Aura page's Agents section describes reliable, repeatable workflow agents completing sophisticated accounting tasks end to end, with the customer writing the rules in plain English and Aura handling the schedule, the data pulls and the handoffs; approval workflows route journal entries, sync events and agent proposals with materiality tiers, and close management runs task checklists across the period. Plain English rule authoring makes this a composable orchestration surface rather than a fixed pipeline.

SourceRillet, rillet.com/product/aura-ai and rillet.com/product/security-and-permissionsread 2026-09-07

Aura answers from a maintained structure: it runs on the customer's live general ledger and company configuration feeding an intelligence layer, answers natural language questions about vendors, accounts, balances or any cut of the data with numbers sourced from the GL, and works against real ledger data rather than a stale copy. The ledger is a persistent, structured, continuously maintained store of the customer's financial record that outlives any run and stays queryable.

SourceRillet, rillet.com/product/aura-airead 2026-09-07

Oversight is stated as product commitments: zero auto-posts without approval, every agent action proposed to a human with nothing posting until approved, approval workflows tiered to materiality with configurable dollar thresholds such as fifty thousand to the Controller and five hundred thousand to the CFO, AI agent proposals routed through the same workflows as journal entries and sync events, and segregation of duties in which agents prepare and humans authorize under role-based access.

The Aura page adds approve and edit on every step. The customer configures that gate, and it sits on the action that matters, posting to the general ledger.

SourceRillet, rillet.com/product/security-and-permissions and rillet.com/product/aura-airead 2026-09-07

Rillet documents attestation and customer-facing controls on one page. Attestation: SOC 1 and SOC 2 Type II with annual third-party audits and reports available for review, AES-256 at rest, TLS 1.2+ in transit, continuous monitoring with penetration tests, and GDPR commitment.

Controls: enterprise SSO via Google, Microsoft Entra and Rippling, three-tier role-based access (Owner, Admin, Accountant plus view-only) with no seat limits and immediate permission changes under full audit history, segregation of duties enforced for AI agents as for people, period locking and per-entity subledger locks. The same page also documents audit logging.

SourceRillet, rillet.com/product/security-and-permissionsread 2026-09-07

One hundred percent of AI actions are logged, with every query, proposal, approval and rejection recorded with timestamp and reasoning, every read access logged with full attribution, every permission change under full audit history, and every period lock and unlock logged with reason and timestamp. That reconstructs what an agent did, why, and who authorized it; the Aura page's audit trail by default and the accrual agents' plain-language rationale sit on the same surface.

SourceRillet, rillet.com/product/security-and-permissions and rillet.com/product/aura-airead 2026-09-07

Aura remembers how you work, the product page says, so it answers, acts and reconciles the way the team would, which describes adaptation rather than a memory feature: no named memory construct, no stated scope or lifetime, no per-user or per-workflow recall surface and no deletion path is documented anywhere. The live general ledger and company configuration feeding the intelligence layer are the application's data and the customer's setup rather than memory, and learned working patterns absorbed into behavior are not a memory a buyer can inspect.

SourceRillet, rillet.com/product/aura-airead 2026-09-07

The platform is hosted on AWS, according to the security page, and the info hub describes delivery as a cloud application with no local install. That is one stated hosting arrangement with no region list, no customer environment option and no selection surface. GDPR commitment is a privacy posture, not a residency option.

SourceRillet, rillet.com/product/security-and-permissions and info.rillet.comread 2026-09-07

Rillet ships named specialized agents a customer puts to work per workflow: the Aura page positions a suite of specialized accounting agents covering flux, accruals, reconciliation and revenue, each scoped to one accounting workflow by the security page's own domain-specialized commitment, alongside the Aura assistant and plain-English workflow agents. Each agent does its own job, an accruals agent without the flux agent remains whole, and the agents ship as adoptable units inside the product rather than as a pricing structure.

SourceRillet, rillet.com/product/aura-ai and rillet.com/product/security-and-permissionsread 2026-09-07

Three trigger classes are documented. Continuous: the Aura page's Continuous AI section has Aura always on, watching for anomalies, flagging exceptions and proposing accruals as they happen. Scheduled: workflow agents run on schedules Aura handles. Event driven: the info hub describes continuous close as real time architecture in which every transaction is fully processed the moment it arrives, with continuous bank and processor syncing, so transaction arrival drives the agent chain rather than a batch cycle.

SourceRillet, rillet.com/product/aura-ai and info.rillet.comread 2026-09-07

Neither a model provider nor any customer or admin choice of model is documented. The security page's model change management commitment confirms Rillet manages its own model stack, with swaps flowing through Rillet's change controls, which is neither disclosed routing across providers nor a customer selection surface.

SourceRillet, rillet.com/product/security-and-permissions and rillet.com/product/aura-airead 2026-09-07

Rillet documents two routes for calling the platform from outside. The Aura page's MCP section connects Aura to Salesforce, Snowflake, Claude or any BI tool with one secure surface for reads and writes, so an outside agent or tool can drive Rillet in both directions. The site's Developers link goes to public API documentation at docs.api.rillet.com with a getting-started guide.

SourceRillet, rillet.com/product/aura-ai and docs.api.rillet.comread 2026-09-07

Nothing puts a change to the customer's agents under test. The security page's model change management commitment puts Rillet's own model swaps through IT general controls with testing, approval and rollback under SOX 404, but what is tested there is Rillet's release practice rather than the customer's agents. Anomaly and error detection runs on the books, the customer's data, not the agent, and no harness, scored test set, live agent metric or release gate is documented for the customer.

SourceRillet, rillet.com/product/security-and-permissions and rillet.com/product/aura-airead 2026-09-07

The agents work inside the general ledger and through native integrations, syncing banks and processors and pulling data over connected systems; no browser, desktop or computer control operated by an agent is documented, and no open research per case at run time is described. The anomaly and flux work runs over the customer's own ledger data.

SourceRillet, rillet.com/product/aura-airead 2026-09-07