Recipes, spec sheets, and allergen matrices are some of the most sensitive files a restaurant group has. Honeycomb processes them in isolated containers that are destroyed after every job, uses only zero data retention models, and keeps an audit trail of everything.
job_7f3a · isolatedYour files go into a container built for one job and destroyed when it finishes. Models only see your data through zero data retention agreements. Only the structured result is kept, unless you choose to keep the originals. Everything is logged. If you need it, we can run inside your own cloud or on your own servers instead.
This is the default path for every menu file in Honeycomb Cloud. The original never lands on a shared server, and the place it was processed no longer exists once the job is done.
You upload menus, recipes, or spec sheets over TLS to a signed, single-use link that expires in minutes and only works for your organization.
We start a new container just for that job. It has no shared disk, no access to other customers, and network access only to approved inference endpoints.
The Honeycomb model runs inside the container. Where another model helps, we call it only through providers that contractually retain nothing.
When the job finishes or times out, the container and its storage are wiped. Your original files go with it. A deletion record is written to your audit log.
Only the processed menu remains: items, allergens, diets, and nutrition, in your Honeycomb workspace where your team reviews it before anything goes live.
Most work runs on our own food intelligence model inside the container. When another model helps, we only use providers whose terms say they do not store your inputs or outputs and do not train on them.
Our purpose built model for ingredients, allergens, diets, and nutrition. It runs inside your job's container, so your data never leaves it.
Open models running on Cloudflare's network, reached through a private gateway with logging of prompts and responses turned off.
You decide what happens to source files, per organization, and you can change it at any time.
Source files are destroyed with the container. Only the structured menu stays in Honeycomb.
Useful when new models or new allergen rules should be rerun over the same recipes without asking your team to upload again.
Most groups start in Honeycomb Cloud. If your policies say menu data cannot leave your infrastructure, we can bring Honeycomb to you.
Our managed platform with a disposable container for every processing job. The fastest way to go live and the option most groups choose.
Honeycomb's processing stack deployed into your own AWS, Google Cloud, or Azure account. Menu data never leaves infrastructure you control.
Honeycomb's model and processing pipeline packaged for servers in your own facility, including environments with no internet access.
| Honeycomb Cloud | Your VPC | On premises | |
|---|---|---|---|
| Where menu data is processed | Isolated, disposable containers in Honeycomb Cloud | Your cloud account | Your own servers |
| Who holds the keys | Honeycomb, per organization | You | You |
| Third party model calls | Zero data retention providers only | Your cloud's private model endpoints | None required |
| Time to go live | Days | Weeks | Weeks to months |
| Upgrades | Automatic | Scheduled with your team | Signed packages you install |
| Relative cost | Standard | Higher | Highest |
An append only record your admins can search and export, so your security team can answer who, what, when, and which model, without asking us.
Security is a practice, not a page. We run ongoing red team and blue team exercises against the same systems your data moves through.
Adversarial testing of the product, the upload path, and our AI features, by people whose job is to find the gap first.
Detection and response that assumes something will eventually go wrong, and makes sure we see it fast.
We are working toward SOC 2 with an independent auditor. Until the report is issued, we are happy to walk your security team through our controls and complete your questionnaire.
TLS 1.2 or higher for every connection. Databases, backups, and file storage are encrypted at rest with AES-256.
Every record is scoped to your organization. Queries, caches, files, and API keys cannot cross from one organization to another.
Owners, admins, and editors get only what their role needs. Multi-location publishing requires an organization admin.
Honeycomb staff have no default access to your menu data. Access is requested, time limited, approved, and logged.
Keys for the API and AI assistant connection belong to your organization, are stored hashed, and can be revoked instantly.
Background jobs such as scheduled publishes call our API with short lived, signed requests, so nothing internal is open to the internet.
AI suggestions, uploads, and assistant edits arrive as drafts. A person on your team approves before guests see anything.
Sous and the AI assistant connection are tested against prompt injection, can only act inside your organization, and every action is logged.
Your menus are not used to train shared models. Zero data retention providers are contractually barred from storing or training on them.
Encrypted daily backups with point in time recovery, so a bad day stays a small one.
A written response plan with on call owners. If an incident affects your data, we tell you promptly with what happened and what we did.
Every change is code reviewed and tested before release, dependencies are scanned, and secrets never live in source code.
The vendors that can touch customer data in Honeycomb Cloud. In your VPC or on premises, this list is yours to set.
| Vendor | What for | Data terms |
|---|---|---|
| Cloudflare | Edge network, Workers AI inference, file storage | Zero data retention on Workers AI |
| DigitalOcean | Application hosting and managed database | Encrypted at rest |
| PropelAuth | Sign in, organizations, and roles | Account details only, no menu data |
Into a container created for that one processing job. It has its own temporary disk, cannot see other customers, and can only reach approved model endpoints. When the job ends, the container and your original files are destroyed. Only the structured menu result is saved to your Honeycomb workspace.
Mostly the Honeycomb food intelligence model, which runs inside the job's container. When another model helps, we call it only through Cloudflare Workers AI, a zero data retention provider that does not store inputs or outputs and does not train on them. Every model call is recorded in your audit log with the provider and region.
No. Your menus, recipes, and spec sheets are not used to train shared models, and our zero data retention providers are contractually barred from training on them.
Yes, if you choose to. By default source files are destroyed after processing. If you opt in, they are stored encrypted in storage set aside for your organization, no person can read them, only processing jobs can, and every read is logged. You can set a retention period or delete everything at any time.
Yes. We can deploy into your own AWS, Google Cloud, or Azure account, or onto servers in your facility, including air gapped environments that use only the Honeycomb model. Both options cost more than Honeycomb Cloud because they need significant dedicated engineering time for setup, security review, and upgrades.
Uploads, processing jobs, container start and destruction, every model call with provider and region, every human or service access, menu drafts, approvals and publishes, retention changes, deletion receipts, sign ins, role changes, and API key events. Admins can search and export them.
SOC 2 is in progress. We are working with an independent auditor and are happy to walk your security team through our controls and complete your security questionnaire in the meantime.
Yes. We run ongoing red team exercises against the upload path, processing containers, API, and AI features, including prompt injection hidden inside menu files. Our blue team practice covers monitoring, alerting, and incident response, and every finding is tracked to a fix and retested.
Book a call with our team. We will walk you through our data flows, deployment options, and subprocessor list, and work through your security questionnaire with you.
Book time with our team to walk through data flows, deployment options, and your security questionnaire.