Synthetic

Connection is documented by type with one named route: the homepage connects bank accounts, payroll system, business tools and email, and the legal tier documents the Google Workspace and Gmail integration in API-level detail, reading message content for transaction categorization and reconciliation under Google's Limited Use policy, plus bank and card feeds and accounting system data among connected Financial Information. Google Workspace is the only integration named on a vendor surface, no connector catalog exists, and the work products flow out by export rather than write-back.

SourceSynthetic, synthetic.ai and synthetic.ai/privacy-policyread 2026-09-07

One fixed loop run by the vendor's agent: connect everything, Synthetic reads the business context, reconciles, asks clarifying questions and produces accrual-basis books, with the legal tier adding that the agentic systems initiate workflows and provide workflow and task management. The customer composes nothing; there is no builder, no runtime and no configurable sequence.

SourceSynthetic, synthetic.ai and synthetic.ai/privacy-policyread 2026-09-07

Grounding is described only at a high level. Synthetic goes in and reads the context of the business from connected banks, payroll, business tools and email, the legal tier adds chart of accounts, corporate ownership structure and related documentation among connected data with Gmail content read for categorization, and the accrual books it maintains are cumulative, each period building on the last. No query surface, corpus description or retrieval mechanism is published.

SourceSynthetic, synthetic.ai and synthetic.ai/privacy-policyread 2026-09-07

The clarifying-questions channel is the one documented human mechanism, and the legal tier draws the rest of the line explicitly: the homepage has Synthetic coming back with sharp clarifying questions on ambiguous context, agent-initiated consultation before proceeding, while the privacy policy's Agentic AI section states the systems categorize, post and reconcile autonomously without routine human review by Synthetic, that personnel do not independently audit, review or verify the correctness of outputs, and that the customer is responsible for deciding whether to rely on outputs at their own risk.

The agent consults on uncertainty but there is no approval gate on posting; the customer only decides afterwards whether to rely on outputs. A rights clause offers human review where law requires it, as a legal right rather than a feature of the product.

SourceSynthetic, synthetic.ai and synthetic.ai/privacy-policyread 2026-09-07

Nothing on the site or in the legal tier documents an attestation or access controls: no certification or attestation of any kind, no SSO, no roles or permission model, and the privacy policy's security section offers only commercially reasonable safeguards boilerplate with an express disclaimer that security cannot be guaranteed. The company is waitlist-stage, with a registered office at an apartment address. No security program for general availability is published.

SourceSynthetic, synthetic.ai and synthetic.ai/privacy-policyread 2026-09-07

Output visibility, with no record of what the agent did. The homepage gives full access to financials and the insights behind them, and the legal tier has the Services preparing management reports and dashboards and generating derived artifacts, categorizations, reconciliations, anomaly flags and suggested entries. No log of agent actions, no trace of what was posted and why, and no customer-openable run record is documented, and the privacy policy states Synthetic personnel do not review outputs, so there is no vendor-side QA record either.

SourceSynthetic, synthetic.ai and synthetic.ai/privacy-policyread 2026-09-07

No memory layer is documented anywhere. The ongoing reconciled ledger is the application's own accounting data; the legal tier's Gmail section states message content is generally processed in memory without storage beyond the feature, retained only case by case; remembering preferences in the policy is cookie language; and the one learning clause, improving models that serve only the specific user whose data is processed, is learning absorbed per user rather than agent memory.

SourceSynthetic, synthetic.ai/privacy-policyread 2026-09-07

One stated hosting country and nothing to choose. The privacy policy says visitors and customers connect through and to servers located in the United States, with information maintained on web servers and systems within the United States, and international transfers framed as consent to US processing. There is no region list, no customer environment and no selection surface.

SourceSynthetic, synthetic.ai/privacy-policyread 2026-09-07

One single-purpose bookkeeping agent for one vertical, software startups, with no library, catalog, template or second adoptable unit documented anywhere.

SourceSynthetic, synthetic.airead 2026-09-07

Agent-initiated operation is documented in the legal tier: the agentic systems initiate workflows and communicate with the customer where appropriate, requesting clarification or additional documentation, without routine human review, and the homepage's connect-then-come-back-when-done model describes an ongoing service working over arriving data. No customer-set schedule, no connected-system event classes, no webhook and no channel beyond the application and its outreach is documented.

SourceSynthetic, synthetic.ai/privacy-policy and synthetic.airead 2026-09-07

The vendor runs its own stack, described on the site as a new kind of AI with tools built specifically for the job, and the legal tier confirms unnamed third-party cloud infrastructure supporting our AI models with a per-user-only training restriction on Gmail data; no model provider is named and no customer or admin choice of model is documented anywhere.

SourceSynthetic, synthetic.ai and synthetic.ai/privacy-policyread 2026-09-07

No API, SDK, MCP surface or developer documentation exists anywhere. The embeddable Stripe-like service sits under the site's own Our Aspiration framing and the release's long-term vision, so it is a plan rather than a product; the consent clause about authorizing a third-party application to access a Synthetic account is template boilerplate rather than a documented interface, and export of the books is a download, not a callable surface.

SourceSynthetic, synthetic.ai and synthetic.ai/privacy-policyread 2026-09-07

No harness, test set, readable result, release gate or live metric on the agent is published for customers. Working with design partners to validate reliability, and the privacy policy's quality assurance and testing language, describe the vendor's own development practice, and the founder's public standard, that nothing releases until it is more reliable than a human bookkeeper, is the vendor's release criterion rather than something the customer can run.

SourceSynthetic, synthetic.ai and synthetic.ai/privacy-policyread 2026-09-07

The agent works through account connections and reads email content via the Google Workspace APIs rather than driving a browser or operating a computer; no interface control and no open research per case at run time is documented.

SourceSynthetic, synthetic.ai and synthetic.ai/privacy-policyread 2026-09-07