rally Trust Center Report a security issue ↗

Security posture · Reviewed August 25, 2026

Security should be
inspectable.

This page describes the controls present in Rally today, names important operational boundaries, and provides a direct path for coordinated vulnerability reporting.

help@rallywith.ai ↗ Privacy Policy →

01 / Implemented controls

What the product
enforces today.

These summaries were checked against the current API, Electron, deployment, logging, and CI paths. They are not certification claims.

01

Regional architecture

  • Separate Global and EU API deployments use separate regional databases, object storage, Durable Object namespaces, queues, and encryption keys.
  • Authenticated requests carry a region claim; a mismatch with the serving region is rejected.
  • EU product telemetry has no Global PostHog fallback.
02

Identity & access

  • Protected API routes require a signed bearer session backed by an active, unrevoked server-side session.
  • Workspace access is rechecked against active membership; owner and administrator actions have explicit role checks.
  • Google sign-in uses state validation and PKCE; the browser-to-desktop handoff uses a single-use exchange ticket with a separate desktop proof.
03

Credential protection

  • Connector credentials use AES-GCM authenticated encryption with connection-bound additional data.
  • Credential rows record a key version so regional keys can rotate without losing old reads.
  • Connector secrets are kept out of model prompts and redacted from structured logs by sensitive field name.
04

Desktop boundary

  • Electron renderers run with context isolation and sandboxing enabled and Node.js integration disabled.
  • A narrow preload bridge mediates privileged operations; unexpected navigation and popup creation are denied.
  • Content Security Policy restricts executable and network sources, and media permission checks are limited to trusted Rally windows.
05

Realtime sessions

  • Realtime connections use short-lived, signed, scope-bound tickets and durable one-time replay receipts.
  • The current desktop removes the credential from the URL and presents it through the WebSocket subprotocol.
  • Legacy query-ticket transport remains accepted for older desktop compatibility and is documented rather than hidden.
06

Connected tools

  • Personal and workspace connections are authorized against the current principal, scope, and membership.
  • Generic webhook ingestion verifies signatures, timestamps, replay windows, event type, and deduplication receipts.
  • External tool writes use durable execution receipts; a write caught mid-flight is marked ambiguous and is not automatically repeated.
07

Operational safeguards

  • API diagnostics use structured request identifiers and a shared redactor for tokens, secrets, credentials, prompts, message bodies, connector content, file content, and model context.
  • Repository CI runs Gitleaks against full history, and each allowlist exception has a test proving it still rejects real key material.
  • Public security reporting is open; an account, NDA, or document request is not required.

02 / Open resources

No lock icons.
No access request.

These are the public documents that apply to Rally. Internal policies and certifications that are not publicly established are intentionally not represented here.

Privacy & dataPrivacy Policy↗ Contractual privacyData Processing Addendum↗ Service providersSub-processor List↗ Service termsTerms of Service↗ Product conductAcceptable Use Policy↗ Website dataCookie & Tracking Notice↗

03 / Service providers

The systems behind
the system.

The repository currently configures these providers for hosting, inference, telemetry, email, transcription, or public-web retrieval. Customer-directed integrations are documented separately in the full list.

Infrastructure

Cloudflare

Regional compute, databases, storage, coordination, queues, delivery, and network security.

AI routing

OpenRouter

Routes bounded model requests to the configured inference providers.

Model inference

xAI · Google · Anthropic

Configured primary, structured, classification, and fallback inference paths.

Transcription

Microsoft Azure

Optional Azure OpenAI real-time dictation and transcription when enabled.

Global telemetry

PostHog

Global product analytics and operational delivery; EU workspaces are not exported to this path.

Email

Resend

Transactional messages, invitations, notifications, and inbound-email routing.

Public web

Firecrawl

User-requested public web search and retrieval when the capability is configured.

Read the complete Sub-processor List, processing purposes, and locations →

04 / FAQ

The useful
fine print.

Is this Trust Center gated?＋

No. Every resource linked here is public. Rally does not require an account, NDA, email submission, or access request to read this page.

Does Rally claim a security certification?＋

No certification is claimed on this page. We list controls that can be tied to the current implementation and will add independent assurance only when it is complete and current.

What does “Global or EU” mean?＋

Rally operates separate Global and EU application architectures with separate regional storage and keys. Network delivery and customer-selected service providers can process data outside the selected architecture, as explained in the Privacy Policy and Sub-processor List.

How are desktop sessions stored?＋

The default desktop mode encrypts stored session material with an application-wide AES-GCM key. Optional Lockdown mode rewraps session material with Electron safeStorage and the operating-system keystore. We do not describe the default mode as device-bound storage.

How do I report a vulnerability?＋

Email help@rallywith.ai with the subject “Security report.” Include the affected surface, impact, reproducible steps, and a safe proof of concept, but do not send credentials or full customer content.

05 / Coordinated reporting

Found something?
Tell us plainly.

Use accounts and workspaces you own or have permission to test. Do not access another person’s data, disrupt service, use social engineering, or run destructive or high-volume testing. Stop if you encounter sensitive information.

Email a security report ↗