Skip to content
AgentLane

Data Processing Agreement

Pursuant to Article 28 of the General Data Protection Regulation (GDPR)

DPA: pending legal finalization. Version 2026-10-12-draft.

Requires review and approval by qualified legal counsel in the relevant jurisdictions before being presented as final contractual/legal documentation.

This draft is filled in from how the AgentLane platform actually works (see the Compliance Center), so counsel reviews facts rather than a blank template. It is not a signable agreement. Do not send it to a partner as final. Items marked [ ] are open questions for counsel. The platform refuses to record any partner DPA as executed until this template is approved.

This DPA is entered into between the agency partner identified in an AgentLane order or signup flow ("Agency", "you") and AgentLane Ltd ("AgentLane", "we"), and forms part of the agreement governing the Agency's use of the AgentLane platform (the "Agreement"). Where the Agreement and this DPA conflict on data-protection matters, this DPA controls.

0. Roles

The platform is resold white-label: an Agency serves many businesses ("End Clients"), each of whose customers ("End Customers") call or text that business.

Party Role for End Customer data (as designed) Status
End Client (the business) Controller [ ] Confirm
Agency Processor acting for the End Client [ ] Confirm. The Agency may instead be a controller or joint controller for its own purposes.
AgentLane Sub-processor acting for the Agency (processor where an End Client contracts with AgentLane directly) [ ] Confirm
AgentLane Controller for Agency account, billing and website-visitor data [ ] Confirm

The platform records the confirmed roles per End Client in the Compliance Center (processing.agencyRole, processing.agentlaneRole). Until they are confirmed, the client is shown as not ready for an EU pilot.

[ ] Decide whether a single back-to-back structure (End Client ↔ Agency DPA, Agency ↔ AgentLane sub-processing DPA) is used, and provide the Agency's End Client DPA template.

1. Subject matter, term, personal data processed, and categories of data subjects

(1) Subject matter. AgentLane processes personal data only to provide the platform: agency and client account administration, missed-call detection and SMS conversations run by automations, AI-generated replies, appointment booking integrations, support chat, billing, data export and deletion.

(2) Term. For the term of the Agreement, then subject to §11.

(3) Categories of personal data.

  • Agency and End Client users: name, email, role, login and audit events.
  • End Customers: phone number, SMS message content, missed-call metadata (calling number, called number, call status, timestamps), name and appointment details where given, outbound-contact consent and do-not-contact records, AI-voice-call metadata (call id, direction, timings, duration, outcome), and — only where the End Client has enabled transcript mode — the transcript text of an AI voice call.
  • Voice call records (voice_calls): Twilio CallSid, the partner id and client id the call belongs to, call status and timestamps, duration, and the caller's number masked to its last 4 digits only. No audio and no transcript is stored in this record. Retained 90 days by default from the call's start time, with a per-tenant override of 1–3650 days, purged nightly by PurgeVoiceCallsJob. The aggregated voice_minutes usage event for the same call is retained as usage/billing history even after the call record is purged (§11).
  • Support chat messages written by Agency and End Client users.
  • Billing contact data for Agencies (card data is handled by Stripe, not stored by AgentLane).

AgentLane's AI voice runtime was deployed on 2026-09-20 and is in limited internal testing against a single AgentLane-operated test client; no End Client traffic has reached it (§12). It streams call audio to Twilio and to the speech providers Twilio orchestrates on its behalf, for the duration of a call, and does not record call audio in the current configuration. Those speech providers are engaged only for those internal test calls and are therefore not yet listed in §2; they will be added there before the runtime is enabled for any End Client. Call transcripts are stored only where the End Client has enabled transcript mode, which is off by default and obliges the runtime to tell the caller at the start of the call; otherwise only call metadata is retained. Should AgentLane introduce call recording, it will be off by default, enabled per End Client by written instruction, announced to the caller at the start of the call, subject to a defined retention period, and notified to the Agency in advance under §6. [ ] Counsel to confirm this section and §12 against the runtime as shipped before the DPA is presented for signature.

(4) Categories of data subjects. Agency users; End Client users; End Customers who call or text an End Client; website visitors who submit forms.

(5) Special categories. The platform is not designed for special-category data, but End Customers of clinics may volunteer health information in messages. [ ] Counsel to advise on Article 9 handling for dental/aesthetics End Clients.

2. Sub-processors and international transfers

AgentLane's platform runs on one server in Frankfurt am Main, Germany. This comes from the production host's IP registration and geolocation, not the hosting contract. [ ] Confirm the provider (IP registered to xTom GmbH), the data-centre location and whether provider support staff access the server from outside the EEA.

The application, the PostgreSQL database, the job queue, object storage and backups are all on that Frankfurt server. AI inference is the exception: it runs on Google Cloud's Vertex AI global endpoint, which routes each request to a Google-selected region anywhere in the world. Message content sent for an AI reply must therefore be assumed to be transferred outside the EEA. See the Google Cloud row below and docs/vertex-region-decision-2026-09-22.md for why no EU region is used.

Sub-processor Purpose Data Location / transfer Transfer mechanism
VPS hosting provider (xTom GmbH per IP registration) Hosts the application, PostgreSQL, Redis, MinIO object storage and Keycloak All platform data, including backups Germany (Frankfurt) — to confirm [ ] Confirm
Twilio Phone numbers, missed-call callbacks, SMS Phone numbers, SMS content, call metadata US-headquartered; no regional option configured [ ] Confirm. Usually contracted by the Agency or End Client directly.
Google Cloud EMEA Ltd / Google LLC (Vertex AI) AI replies (automations, voice, support chat) — the active default AI provider for all tenants Message content, business context Vertex AI global endpoint (GOOGLE_VERTEX_LOCATION=global) — Google routes each request to a region of its own choosing, worldwide. No EEA residency; an international transfer must be assumed. [ ] Confirm. Google Cloud Data Processing Addendum plus SCCs proposed; not yet executed or verified.
Anthropic Emergency circuit-breaker fallback only, on the platform key, if Vertex is unavailable. Not used for tenant BYOK — Anthropic BYOK is retired Message content, business context US-headquartered [ ] Confirm. Zero data retention is not confirmed.
OpenAI AI replies (alternate provider) Message content, business context US-headquartered [ ] Confirm. Zero data retention is not confirmed.
Cal.com Booking integration Name, contact, appointment time Depends on the End Client's account Contracted by the End Client
Resend Transactional and automation email Name, email, email content US-headquartered [ ] Confirm
Stripe Agency billing Agency billing contacts US-headquartered [ ] Confirm
Slack Internal operational alerts Agency names in alerts (no conversation content) US-headquartered [ ] Confirm

PostgreSQL, Redis, MinIO and Keycloak are self-hosted software on the server above, not separate sub-processors. Automations run inside the AgentLane API on that same server. n8n is no longer used, and neither Retell nor Vapi receives any data.

Where the Agency or End Client supplies its own provider credentials (BYOK), that provider acts under the Agency's or End Client's own contract.

[ ] For each US provider: the SCC module (2021/914), DPF certification, or other Article 46 mechanism; a transfer impact assessment (docs/tia-us-ai-transfers.md currently asserts SCCs and DPF without evidence and must be re-verified).

[ ] For Google Cloud specifically: because the global endpoint does not pin a region, the transfer impact assessment cannot be scoped to one country. Confirm the Google Cloud Data Processing Addendum and the SCC module that covers it, and whether Google's own regional controls can be reconciled with the model availability recorded in docs/vertex-region-decision-2026-09-22.md.

Vertex AI activation notice. This notice was given, as required by §6, before GOOGLE_VERTEX_PROJECT, GOOGLE_VERTEX_LOCATION and the service-account credentials were configured. As of the 2026-09-22 deploy, those variables are all set in production and Google Cloud (Vertex AI) is the active default AI processor for all tenants (Gemini 3.1 Flash-Lite). GOOGLE_VERTEX_LOCATION is being set to global: the Gemini 3.x models AgentLane runs are served only by Vertex's global endpoint, and the EU single-region endpoints (europe-west4, europe-west1) serve only the older Gemini 2.5 family. This section was corrected before that value was applied, so this DPA never describes narrower processing than the platform performs. AgentLane therefore makes no EEA-residency commitment for AI inference, and this activation notice is also notice of that transfer. Claude-on-Vertex exists in code but is gated off. Anthropic BYOK is retired — no inference path reads a tenant's own Anthropic key — and Anthropic is reachable only through the emergency circuit-breaker fallback on the platform key if Vertex is unavailable.

The machine-readable version of this table is packages/shared/src/compliance/subprocessors.ts.

3. Technical and organisational measures

Measures in place today:

  • Tenant isolation: every Agency and End Client query is scoped server-side, and a CLIENT user is pinned to its own client.
  • Provider credentials encrypted with AES-256-GCM envelope encryption at the application level.
  • TLS on all public endpoints.
  • Application audit log of partner, client and admin actions.
  • Role-based access, including view-only agency moderators. MFA is available but optional.
  • Automated retention and deletion jobs, each run logged (§11).
  • Firewall exposing only SSH, HTTP and HTTPS.

Known gaps disclosed to counsel: database backups are stored on the same disk as the database, with no offsite copy and no recorded restore test; object storage (MinIO) is not backed up; disk-level encryption of the host is not confirmed; MFA is not enforced.

[ ] Agree the TOMs annex wording.

4. Data subjects' rights

(1) AgentLane assists the Agency, and through it the End Client, with data subject requests. Available tools: portal data export (ZIP of JSON/CSV), account deletion requests, a per-client do-not-contact list, and SMS STOP handling.

(2) Requests received directly from an End Customer are passed to the responsible Agency/End Client without undue delay. AgentLane does not answer them on its own authority unless instructed. Identity checks are proportionate. Government ID is not requested. Verification happens by email, not SMS.

5. Further duties of AgentLane

Establishment and representative. AgentLane Ltd is a company registered in England and Wales (No. 17400209). It is not established in the EU/EEA. [ ] Counsel to determine whether an Article 27 EU representative is required, and name it here if so.

Data protection contact. No DPO has been appointed. Contact: contact@agentlane.agency. [ ] Confirm whether Article 37 applies.

Confidentiality. Personnel with access to personal data are bound by confidentiality obligations.

Cooperation with supervisory authorities. AgentLane cooperates with authority requests and informs the Agency promptly, where permitted.

6. Sub-processor changes

AgentLane will notify the Agency of any new sub-processor before it processes the Agency's data, giving the Agency an opportunity to object. [ ] Agree the notice period (30 days proposed) and the remedy if the Agency objects.

7. Audits

On written request, AgentLane provides information showing compliance with this DPA, including evidence of the measures in §3. On-site audits are not offered at the current scale. [ ] Revisit if an Agency's obligations require third-party audit rights.

8. Assistance and personal data breaches

AgentLane assists with Articles 32–36 GDPR. It notifies the Agency of a personal data breach affecting the Agency's data without undue delay after becoming aware of it. [ ] Agree a maximum notification period and the notification content. No incident-response runbook has been formally approved yet (see docs/policies/security-incident-policy.md).

9. Instructions

AgentLane processes personal data only on documented instructions: the Agreement, this DPA, and the Agency's and End Client's configuration of the platform, including retention periods, AI disclosure wording and outbound policy. The only exception is where EU or Member State law requires otherwise; AgentLane will inform the Agency of that requirement unless the law prohibits it.

10. Liability

As set out in the Agreement, consistent with Article 82 GDPR. [ ] Counsel to align with the Agreement's liability cap.

11. Retention, deletion and return of personal data

  • Export. The Agency and each End Client can export their data from the portal at any time. Export archives are deleted automatically after 7 days.
  • Conversation retention. Conversations are deleted 12 months after last activity by default. The Agency or End Client can set a period of 1–3650 days, and the period cannot be unlimited.
  • Voice call record retention. Per-call voice records (voice_calls — CallSid, party ids, status, timestamps, duration, and the caller's number masked to its last 4 digits; no audio, no transcript) are deleted 90 days after the call by default. The Agency or End Client can set a period of 1–3650 days. Purged nightly by PurgeVoiceCallsJob. The aggregated voice_minutes usage event for the same call is deliberately retained afterward as usage/billing history.
  • Closure. An approved deletion request disables the account immediately and schedules a permanent purge 30 days later. The purge deletes conversations, End Customer records in the legacy home-services tables linked to that client, provider credentials, consent and do-not-contact records, and export archives, and it anonymises user records. Billing records are kept as required by law.
  • Backups. Deleted data remains in database backups until they rotate out (14 days).
  • Phone numbers. Numbers stay in the Twilio account that holds them. AgentLane does not transfer or release numbers held in an Agency or End Client account.
  • Known gaps. Legacy home-services data not linked to a client or agency is not purged automatically. Retention for support chat, webhook delivery logs and the audit log currently runs in dry-run mode.

The full retention register is published in the Compliance Center and packages/shared/src/compliance/retention.ts.

12. AI transparency and outbound calling (technical annex)

  • AI disclosure. Every End Client's voice configuration has a localized AI-identity disclosure enabled by default. The call-start contract (buildVoiceCallStartPlan) places it ahead of the business prompt with barge-in disabled while it plays. Only AgentLane can disable it, and only with a recorded reason. The platform records disclosure evidence (status, locale, wording version, call id) without audio. As designed, the runtime plays the disclosure as the first utterance of every call and does not act on anything said over it. The enforceable guarantee is that AgentLane does not stop playing the disclosure and does not use what a caller says during it — it cannot prevent a caller from speaking. The per-call disclosure record (wording version, locale, spoken-at timestamp) is the transparency evidence, and it contains no audio. The voice runtime was deployed on 2026-09-20 and is in limited internal testing against a single AgentLane-operated test client; the disclosure has been played only in those internal test calls, and no End Client traffic has reached it. [ ] Counsel to confirm the wording satisfies Article 50(1) EU AI Act for each locale and use case, and to confirm the transcript notice wording where transcript mode is enabled.
  • Outbound calling. Disabled by default. No dialer exists. Before any outbound call is permitted, the platform requires: an admin-recorded legal approval, jurisdiction, calling entity, allowed purposes, a recorded consent (where required), no do-not-contact match, allowed calling hours in the client's timezone, and an active disclosure. [ ] Counsel to set the applicable jurisdiction-specific consumer-protection, do-not-call, and marketing-consent rules that the policy must encode.

Requesting the current draft: email contact@agentlane.agency. A signable version will be published only after counsel approval.

Data Processing Agreement — AgentLane