Can you own it?

Data governance - retention, ZDR & training

Together AI's headline governance posture is zero data retention by default, and this session it was read in the binding documents, not a marketing page. The Privacy Policy states that inputs and outputs are not stored by default, and that under ZDR content is not stored, retained, or used for model training or product improvement except as needed to provide the Services - aside from temporary caching for performance. The Terms of Service add that Together does not use your data to train its models without your explicit opt-in and consent, so training is opt-in and off by default, gated behind an org-admin Privacy & Security setting.

The caveat is provenance of assurance, not policy: this is Together's own self-attestation in its binding terms - strong, but not independently audited - and no public sub-processor list was found. For a regulated deployment, contractualise the no-retention behaviour in the DPA rather than relying on the default terms.

Data & IP ownership

The Terms of Service settle ownership directly: as between you and Company, you exclusively own all right, title and interest in Your Content and Output, granting Together only a limited license to access, host and operate that content solely to provide the Services. Together separately owns its own Company IP and Usage Data. Combined with ZDR-by-default from the Privacy Policy - nothing retained to improve the service unless an org admin opts in - the practical posture is that your data and the outputs derived from it stay yours. The one gap is disclosure: the sub-processor list is not published, so for a regulated deployment pin sub-processor terms in the DPA.

Compliance & attestations

Together announced SOC 2 Type II through an independent, multi-month audit (per its own blog - the announcement was read, the SOC 2 report itself was not obtained), and its Data Processing Addendum - read this session - binds it to the EU Standard Contractual Clauses for EU personal-data transfers, satisfying the GDPR DPA requirement. HIPAA/BAA support is indicated in the security docs via encryption, audit logging and business associate agreements rather than shown as a standalone attestation, and ISO 27001 / 27701 are unconfirmed - the DPA aligns risk management to ISO 27005 / NIST 800-37, which is not a 27001 certification. For healthcare or ISO-dependent procurement, request the current attestation letters and the SOC 2 report directly.

Data residency & jurisdiction

Together's support knowledge base (read this session) confirms US and EU data centres (including Sweden GPU capacity) with dedicated, region-pinned endpoints; the DPA confirms EU transfers run under SCCs with default routing through North America. The constraint is commercial rather than technical: EU residency and dedicated/region-pinned endpoints are gated to the Scale and Enterprise tiers, not available on the self-serve serverless plan. As a US-headquartered company, Together also carries standing US CLOUD Act exposure even for EU-hosted workloads - a jurisdictional fact no region setting removes.

Security controls

The security docs and the SOC 2 announcement (both read this session) describe encryption in transit and at rest, audit logging, MFA/RBAC, continuous monitoring and regular penetration testing, and the DPA requires encryption of controller data in transit and storage. No independently reviewed penetration-test report or the SOC 2 report itself was obtained, so this is assessed as solid-but-partial: the day-to-day controls are documented, but the deepest independent assurance is not visible.

Pricing & cost model

Pricing is mixed: per-token serverless rates for the shared catalogue plus dedicated GPU endpoints on Scale/Enterprise. Together's pricing page lists the models, and indicative serverless rates corroborated by independent third-party price analysis - gpt-oss-120B ~$0.15/$0.60 per M in/out, Qwen3-235B ~$0.20/$0.60, DeepSeek-V3 ~$1.25/M - are approximate and move frequently (the live pricing page was not read from source this session), so treat them as an order-of-magnitude guide and confirm live rates before committing.

Reliability posture

No incidents surfaced in research, but no public uptime SLA, status page or failover documentation was found or independently verified this session. Reliability is therefore assessed conservatively: adequate on available evidence, but without an observable status/SLA record to lean on for availability-critical workloads.