Rithmo Security & AI Data Privacy Whitepaper

Last updated: July 2026

Rithmo maintains a record of the decisions an organization makes. It reads the systems a company already uses (meetings, messaging, email, CRM, and documents) and reconciles them into a single account of what was decided, who owns it, whether it still holds, and what evidence supports it.

Doing that means processing some of the most sensitive material a company has. Decisions are made in conversations about customers, strategy, pricing, personnel, legal exposure, and security. Rithmo is built on the principle that a system can hold that record only if the data behind it is processed through clear boundaries, strong controls, and transparent vendor practices.

Rithmo's security and privacy model rests on six commitments:

Rithmo is not designed to avoid AI. It is designed to make AI use controlled, minimized, explainable, customer-scoped, and auditable.

This whitepaper explains how Rithmo protects customer data across its deployment models, connected sources, data lifecycle, AI processing model, retention practices, and security controls.

It is intended for customer security, privacy, legal, procurement, and IT teams evaluating Rithmo for business use.

This document covers:

• deployment models and where customer data resides;
• customer data categories processed by Rithmo;
• how connected sources are read and reconciled into the decision record;
• AI and external provider processing;
• tenant isolation and cross-customer contamination prevention;
• customer controls and administrative governance;
• access control, encryption, logging, retention, and deletion; and
• subprocessor and external service transparency.

This whitepaper should be read together with Rithmo's Privacy Policy, customer agreement, Data Processing Addendum, and subprocessor disclosures.

Where Rithmo runs determines who holds the data, and it is the single most consequential security decision a customer makes about the product. Rithmo supports two models.

3.1 Customer-Hosted (Self-Managed)

Rithmo runs inside the customer's own cloud environment. Connected-source content (messages, transcripts, documents, CRM records) is read, processed, and stored within infrastructure the customer controls. It does not transit or reside in a Rithmo-operated platform.

In this model:

• the customer owns the storage, the encryption keys, and the network boundary;
• the customer's existing controls, logging, and compliance posture apply directly;
• there is no copy of customer business content held by Rithmo to audit, subpoena, or breach; and
• the review surface for a vendor security assessment narrows considerably, because the vendor is not holding the data.

Customer-hosted deployment does not eliminate the need for a security review. External AI providers may still be called from within the customer environment, subject to the controls in Section 8, and the customer remains responsible for configuring the deployment in line with their own policies.

3.2 Rithmo-Managed

Rithmo operates the service and processes customer content in Rithmo-controlled infrastructure. This is the appropriate model for organizations that prefer not to run the software themselves.

In this model the controls described throughout this document (encryption, access control, tenant isolation, retention, logging, and incident response) are operated by Rithmo, and Rithmo acts as a processor of customer data under the applicable customer agreement and Data Processing Addendum.

3.3 Choosing a Model

Customer-hosted deployment is the stronger privacy posture and the model Rithmo recommends for organizations handling regulated, confidential, or otherwise high-sensitivity material. Rithmo-managed deployment remains fully supported and is a legitimate choice where operational simplicity matters more than holding the data in-house.

Both models are covered by the same commitments in Section 1. Where a control in this document applies only to one model, it says so.

4.1 Decision Content Is Sensitive by Default

Rithmo treats the material it reads (messages, transcripts, documents, CRM records, calendar data) and everything derived from it, including the decision record itself, as sensitive customer business content. It is not treated as generic usage telemetry.

4.2 AI Is Governed by Product Boundaries

Rithmo may use AI providers to support transcription, extraction, classification, summarization, embeddings, and reconciliation. These providers are used through defined product pathways, not unmanaged employee workflows or ad hoc data movement.

4.3 Customer Data Remains Customer-Scoped

Customer-specific content, decision records, embeddings, retrieval context, prompts, and derived intelligence remain scoped to the relevant customer, workspace, team, or authorized user context.

4.4 Sensitive Content Is Minimized Before External AI Processing

Rithmo's AI processing model sends only the minimum data required for a given task. Where content is needed for an AI feature, Rithmo uses bounded context, redaction, pseudonymization, identifier separation, and local rehydration to reduce unnecessary exposure.

4.5 The Record Is Evidence-Backed, Not Model-Asserted

A decision in the record is not a model's opinion. Each entry carries the evidence it was formed from and the provenance chain behind it, so a customer can inspect why Rithmo believes something was decided, and contest it. AI assists in forming the record; it does not become the record unchallenged.

Rithmo has two customer-facing surfaces: connectors that read the systems a customer already uses, and an application where the decision record is reviewed, contested, and acted on. Rithmo also exposes the record to agents and other software through a controlled read interface.

Behind these surfaces, Rithmo coordinates ingestion, reconciliation, the decision record, integrations, and customer-facing updates through a central application layer.

The key architectural principle is controlled data flow. Product behaviour is routed through defined application, policy, runtime, integration, storage, worker, and provider pathways, which reduces the risk of inconsistent data handling or uncontrolled automation.

Rithmo reads customer systems through connections the customer authorizes and can revoke.

• Scoped access. Connections use standard OAuth or equivalent mechanisms, requesting the narrowest scopes the feature requires.
• Customer-revocable. A customer can withdraw access at any time from their own administrative tooling, without depending on Rithmo to act.
• Read-oriented. Rithmo's core function is to read and reconcile. Where a feature writes back to a connected system (posting to a channel, updating a record) that action is routed through the execution controls in Section 9 and is subject to customer configuration.
• Bounded by source permissions. Rithmo does not see more than the connection grants. Content a connected account cannot access is not available to Rithmo through that connection.

Rithmo uses AI and model-based services to support capabilities such as transcription, extraction of decisions and commitments from conversation, classification of state transitions, semantic similarity and embeddings, summarization, and reconciliation across sources.

AI processing is governed by the following controls.

8.1 Registered Product Purpose

Each external AI processing pathway is tied to a defined product purpose, such as extraction, classification, transcription, embedding, or summarization.

8.2 Minimum Necessary Data

Rithmo sends only the information needed for the specific feature. A task requiring a short excerpt does not receive a full transcript or document.

8.3 No General-Purpose Model Training

Rithmo does not use customer content to train general-purpose AI models, and requires external AI providers not to train their models on customer content unless the customer expressly authorizes that use.

8.4 Contractual Vendor Controls

External AI, transcription, speech, infrastructure, and workflow providers are governed through contractual and technical controls appropriate to the type of service and data involved.

8.5 Content-Free Operational Logging

AI usage logging captures operational metadata (provider, model, task type, latency, success or failure state, token usage where applicable) without storing raw prompts, raw content, or full model responses in routine operational logs.

8.6 Deployment-Dependent Provider Access

Under customer-hosted deployment, external AI provider calls originate from within the customer's environment. Customers may restrict which providers are reachable, supply their own provider credentials, or disable externally-served AI features entirely, accepting the corresponding reduction in capability.

Some of Rithmo's core capabilities require processing conversation and document content. Rithmo's security model does not depend on pretending that content is never used. It governs that use carefully.

Documents often contain the most sensitive material in a connected environment. Rithmo treats uploaded or linked documents as high-sensitivity customer content.

Document processing is governed by explicit feature enablement, access controls tied to customer permissions, encryption for stored sensitive artifacts where supported, external AI processing only where the enabled feature requires it, minimization and redaction before external processing where feasible, retention controls aligned to customer configuration, and audit logging that avoids raw document content.

Enterprise customers may restrict document processing, require approved providers, or apply stricter configurations.

Rithmo is a customer-scoped platform. One customer's content is never used to generate customer-facing outputs for another customer.

This applies to prompts, summaries, the decision record, embeddings and retrieval context, caches, learning records, derived intelligence, analytics, and all customer-facing output generation.

The isolation model is built on customer-scoped storage, permission-checked retrieval, scoped prompt context, scoped caches, scoped embeddings, and customer-specific learning boundaries. Under customer-hosted deployment, tenant isolation is additionally enforced by the fact that the deployment serves one customer.

Depending on plan and enabled features, administrators may control:

• the deployment model;
• which sources are connected, and with what scopes;
• whether transcription is enabled;
• which AI-enabled features are active;
• whether document processing is enabled;
• which users and roles may access the decision record and its evidence;
• retention settings for source content, the record, derived intelligence, documents, and logs;
• stricter handling for sensitive categories;
• whether Rithmo may write back to connected systems; and
• whether customer content may be used for optional product-improvement programs.

Customer administrators are responsible for configuring Rithmo in accordance with their organization's internal policies, participant notice requirements, and applicable law. Where Rithmo reads recorded conversations, customers are responsible for meeting any notice or consent obligations that apply to those recordings.

Rithmo uses third-party providers to operate the service and provide customer-enabled features. Provider usage depends on the customer's deployment model, configuration, and enabled features. Under customer-hosted deployment, several categories below are provided by the customer's own infrastructure rather than by Rithmo.

Rithmo maintains subprocessor disclosures identifying providers, processing purposes, and applicable data categories.

14.1 Standard Users

Users may access only the records, evidence, and workflows authorized by the customer's configuration and their role. Access to a decision's evidence follows the permissions of the underlying source material.

14.2 Customer Administrators

Administrators may have broader visibility within their own organization (settings, integrations, reporting, retention controls, and record history) depending on configuration.

14.3 Rithmo Operational Access

Under Rithmo-managed deployment, Rithmo operational access to customer content is restricted to authorized personnel with a legitimate business, support, security, or legal need, governed by least privilege, access controls, and auditability.

Under customer-hosted deployment, Rithmo personnel have no standing access to customer content. Any access for support purposes is initiated and granted by the customer.

15.1 Encryption in Transit

Rithmo uses HTTPS/TLS-based communications for user-facing application traffic and external provider connections.

15.2 Encryption at Rest

Rithmo uses storage-layer encryption for databases, object storage, backups, and infrastructure services where applicable, and application-layer encryption for selected high-sensitivity data classes. Under customer-hosted deployment, encryption at rest is provided by the customer's infrastructure and keys.

15.3 Application-Layer Encryption

Application-layer encryption may be used for high-sensitivity data such as integration and OAuth tokens, document artifacts, uploaded attachments, generated briefs, and other high-sensitivity content.

15.4 Credential and Token Protection

OAuth tokens, session secrets, API keys, and integration credentials are protected through encryption, hashing, limited access, and secure storage practices appropriate to the credential type.

Rithmo uses structured logging and an audit-oriented architecture to support reliability, security, troubleshooting, customer diligence, and incident response.

Operational logs capture request identifiers; customer, organization, or team scope; task type; provider and model information; latency; token count where applicable; success or failure state; fallback usage; feature flag state; redaction status; retention mode; and policy or guardrail outcome.

Rithmo does not store raw content, full prompts, or full model responses in routine operational logs. Elevated debugging or support workflows involving sensitive content are access-controlled, time-limited, and governed by support, security, and customer-authorization controls.

Separately from operational logging, the decision record maintains its own provenance: the evidence each entry was formed from, and the sequence of transitions it has undergone. This is a customer-facing feature, not a log, and is subject to the same access controls as the record itself.

Upon termination or verified deletion request, Rithmo deletes or returns customer data according to the applicable customer agreement, legal obligations, product capabilities, and backup deletion cycles. Under customer-hosted deployment, deletion is performed by the customer within their own environment.

Rithmo's security model supports a structured incident response lifecycle:

1. Detect and triage suspected security events.
2. Investigate using audit trails, logs, provider telemetry, and system metadata.
3. Contain affected systems, credentials, integrations, or workflows.
4. Remediate and validate corrective action.
5. Communicate with affected customers where required by contract, law, or severity.
6. Improve controls based on lessons learned.

Incident notification timelines and contractual commitments are governed by the applicable customer agreement or Data Processing Addendum. Under customer-hosted deployment, incidents confined to the customer's environment are the customer's to detect and respond to; Rithmo supports that response on request.

Rithmo supports enterprise diligence through materials that may include the Privacy Policy; Data Processing Addendum; subprocessor list; AI processing summary; retention summary; security architecture overview; incident response summary; access control and privileged access summary; encryption and key management overview; vendor data protection commitments; deployment and data residency information; and security assessment materials where available.

These materials are intended to help customers evaluate Rithmo's data protection posture without relying on broad marketing claims. Rithmo does not currently hold a SOC 2 attestation; where a customer's process requires one, Rithmo will say so directly rather than imply otherwise.

Rithmo helps organizations hold a reliable record of what they have decided, while preserving control over the sensitive material that record is built from.

The platform's security and privacy posture rests on:

• a deployment model the customer chooses, including running entirely in their own environment;
• controlled ingestion from customer-authorized, customer-revocable connections;
• AI processing through defined provider pathways;
• content and identifier minimization before external processing;
• no general-purpose AI training on customer content;
• an evidence-backed record that can be inspected and contested;
• tenant isolation;
• customer controls;
• retention and deletion practices; and
• auditability.