Trust center
The safest AI action is the one
a person approved.
Security at BudAlly starts with the product design: agents hold no execute rights, every write runs under a named person's credential, and every proposal leaves an audit trail you can export. Then the usual controls on top.
- SOC 2 Type IIPlaceholder: [STATUS · DATE]
- Metrc Validated IntegratorPlaceholder: [STATUS · STATES]
- Pen testPlaceholder: [LAST DATE · FIRM]
- IncidentsChangelog & incident history
01 · Draft-then-approve as a control
Agents can't do anything. They can only propose.
Every agent run produces proposals with evidence, a confidence score, the rules it checked and a precondition hash. If the world changes before approval — a price moved, a manifest was edited — the proposal is superseded and can't be approved. Cutovers and Metrc-affecting changes need two different approvers. There is no “approve all” anywhere in the product.
- Executes under the approver's own POS or Metrc credential
- Role-based access; maker ≠ checker on campaigns and cutovers
- Learned rules are versioned and approved before they apply
- Audit log: who, what, when, evidence, credential — exportable as CSV/JSON
- Proposed
- Reviewed
- Approved
- Revalidating
- Executing
- Executed
- Verified
- Superseded
The controls around it
02 · Data ownership
Yours. Contractually.
Ledger, members, consent proofs, documents export in open formats at any time, on any plan. No resale, no cross-tenant training. Deleting a tenant purges within Placeholder: [N] days with a certificate.
03 · Tenancy & aggregation
k ≥ 8, or it doesn’t show.
Each license is its own tenant row with its own credentials. Network medians need at least 8 contributors and explicit opt-in; Dutchie-connector rows are excluded from all benchmarks under its developer terms.
04 · Model providers
30-day retention or less. No training.
Invoice PDFs and copy drafts go to a model provider under a zero-training, ≤30-day retention agreement. A zero-retention tier is available on Chain Ops. Providers are listed as sub-processors.
05 · Infrastructure
Encrypted, backed up, US-hosted.
TLS 1.2+ in transit, AES-256 at rest, POS and Metrc keys in a vault with per-tenant rows, daily encrypted backups with Placeholder: [N]-day retention, US region: Placeholder: [PROVIDER · REGION]. Sentry for errors, no PII in traces.
06 · Consent & marketing law
Proof, not checkboxes.
Every contact carries channel, timestamp, source and text version. No SMS without your own 10DLC and proven consent. CA financial-incentive notices and NY inducement bans are enforced before a draft exists (TCPA, CAN-SPAM, CCPA — counsel reviewed).
07 · Access
SSO, roles, sessions.
Google and Microsoft SSO, SAML on Chain Ops, MFA enforced for approvers, role-based access down to store and module, session logs, kill switch per connector and per user.
Sub-processors
| Provider | Purpose | Region |
|---|---|---|
| Placeholder: [HOSTING PROVIDER] | Application hosting, database, backups | US |
| Anthropic | Model provider for invoice capture and copy drafts (≤30-day retention, no training) | US |
| Resend | Transactional email (briefings carry titles and counts only) | US |
| Sentry | Error monitoring, PII scrubbed | US |
| Google Maps Platform | Delivery routing (addresses only) | US |
| Headset | Licensed market data (inbound only; no tenant data sent) | US |
Responsible disclosure
Found something? security@budally.com · PGP key on this page · we acknowledge within 2 business days and don't pursue good-faith researchers.
PGP fingerprint: Placeholder: [KEY]
Documents on request
SOC 2 report, pen-test summary, DPA, security questionnaire (CAIQ), business continuity plan.
Request the security packet