AI agents and plugins
Agent platform capability detail, including OpenAI-compatible interfaces and MCP servers.
View AI agents →Private AI infrastructure
Private AI, with a company-controlled interface and a gateway to the models you choose
Give staff one company AI workspace instead of scattered personal accounts. We set up a chat interface that you control (Open WebUI or LibreChat), connect it to the Requesty AI gateway, and route to the models you select, with access, logging, and spend controls scoped in writing.
Private here means company-controlled access and hosting of the interface. It does not mean local model inference: the models run at the providers you pick, so prompts still reach them. Open WebUI, LibreChat, and Requesty are third-party products. Nothing here implies endorsement, partnership, or reseller status, and we do not certify compliance.
The wish list is the low-friction first step. The $99 consultation is a separate, paid, 1-hour live session, priced apart from any assessment, build, or support plan.
Who this is for
- Companies whose staff already use personal AI accounts, and whose leaders want one governed place to do it.
- Teams that want to mix or switch models without rebuilding the interface each time.
- Firms handling client or regulated information that need clear answers on where prompts go, what is logged, and who can see it.
- Operations leads who want per-team usage and spend visibility instead of a pile of individual subscriptions.
Deliverables
What a private AI stack includes
Scope is chosen from this list. You do not need every piece, and Open WebUI and LibreChat are alternatives: we recommend one, not both.
Interface setup
Open WebUI or LibreChat, installed and configured in a cloud account you control or one we host under a service plan. We recommend one after comparing features, sign-in options, and your team's needs.
Gateway and model selection
The interface connects to the Requesty OpenAI-compatible gateway, so it talks to one endpoint instead of each provider. We help you choose the models staff may use and a default for each use case. Availability depends on the gateway and provider.
Access and sign-in
Roles, groups, and single sign-on where the interface edition and configuration support them. We confirm what each option supports before we put it in a scope.
Logging and retention
Decide what is kept and where: chat history in the interface, logs at the gateway, and retention at each model provider. We set what can be set and document the rest.
Spend controls
Budgets, limits, and usage reporting using the controls your gateway plan offers, so cost is visible per team instead of discovered on the invoice.
Runbook and training
An admin runbook, short guidance for staff on what may go in, and training. Optional file upload, document search, or tool connections are scoped separately against your data rules.
Decision
Interface, gateway, model: three separate choices
Most private AI confusion comes from blending these. Each layer has its own owner, its own data handling, and its own bill. You do not need every layer: a hosted workspace may be enough.
Interface
What it is:The chat app staff open. Open WebUI or LibreChat, running in a cloud account or on a server you control.
What to decide:Who may sign in, which features are on, where chat history and uploads are stored, and how they are backed up.
Gateway
What it is:One OpenAI-compatible API between the interface and the model providers. Requesty is a managed service run by Requesty, not something you host.
What to decide:Which models are exposed, routing and fallbacks, logging and retention settings, budgets, and whether to use its EU endpoint.
Model provider
What it is:The company that runs the model, such as OpenAI, Anthropic, or another provider the gateway supports. The model runs on its infrastructure.
What to decide:Which providers and models are allowed, their data and retention terms, and where they process data.
Local inference
What it is:Only if prompts must never leave your own hardware. That is a different project: you run open-weight models yourself.
What to decide:Not part of this stack. If it is a hard requirement, say so on your wish list and we will tell you whether we can scope it.
Not sure you need a private stack? A hosted workspace such as a ChatGPT business plan or Claude for teams has less to run. We compare both before you commit.
Process
How a rollout runs
Small group first, with the data map agreed before anyone pastes real data in.
- Stage 1
Scope and data rules
Agree who uses it, what data may go in, which models are allowed, and which residency or retention needs are real requirements.
- Stage 2
Stand up
Install the interface where you chose, connect it to the gateway, and configure access with test users.
- Stage 3
Set controls
Model list, budgets, logging and retention settings, sign-in, and backups, each checked against the data map.
- Stage 4
Pilot
A small group on real work, with feedback on models, limits, and settings before wider rollout.
- Stage 5
Hand off
Runbook, staff guidance, and training. Then expand teams or models, or add support if we host it.
Timelines are set in a written scope once hosting, sign-in, models, and review turnaround are known. Estimates, not guarantees.
Governance
EU data location: check four places, not one
EU residency is an option here, not a blanket promise, and it is different from self-hosting. An EU gateway endpoint covers the gateway's own processing. Where chats are stored, where the model runs, and who else touches the data are separate questions, so we check each one.
Interface and uploads
We choose regions for the interface, its chat storage, and uploaded files with you, and check any separate databases or storage services. The gateway does not decide those locations.
Gateway and logs
Requesty documents an EU endpoint, and says its prompt and output logging is on by default for a limited period and configurable by policy. We confirm the setting for your account in writing and do not assume it is off.
Model provider processing
Requesty says its EU endpoint does not ensure the model runs in the EU. We select EU-hosted models, enforce gateway region restrictions and an EU-only approved-model list, and verify the selected provider's endpoint and terms.
Backups and subprocessors
Backups, monitoring, email, sign-in, and each vendor's own subprocessors are more places data can sit. We list them in the data map.
Contract and DPA
Requesty makes a data processing agreement available on request. A DPA is a contract, not a setting, and Requesty says it covers Requesty as processor, not the model providers. Providers need their own terms, and your counsel decides if they suffice.
Self-hosting is not residency
Hosting the interface yourself puts the app and its stored data under your control. It does not keep prompts local, because they still go to the gateway and a provider. An EU gateway does not make the interface or the provider EU either.
What a data-location review produces
- A data map covering the interface, gateway, each provider, backups, and subprocessors
- Region and endpoint choices written down per layer
- Retention and logging settings, and who can see the data
- A list of vendor terms and DPAs to request, for your counsel to review
- Gaps we could not confirm, flagged instead of assumed
- A reminder to re-check when a provider, plan, or model changes
This is operational guidance, not legal advice, a compliance certification, or a guarantee that data stays in any region. Everyday Workflows is not a party to Requesty's or any provider's DPA. How we handle your data.
Ask for residency in writing
If EU data location or a DPA matters to you, say so on your wish list. We scope the checks above before choosing the stack, so you find out early if a requirement cannot be met.
Pricing & terms
How the cost works
Three layers, three bills. Our fees are separate from software licenses, gateway usage, and model usage.
- Our setup fees
- Quoted from an agreed written scope. Build work uses 50% to start and 50% after the agreed KPI review. Out-of-scope work goes through a change order with the price stated first.
- Hosting the interface
- If we host it, it runs under license on a monthly service plan quoted from the scope. If it runs in your own cloud account, a one-time setup and a lighter maintenance plan apply instead.
- Interface licensing
- Open WebUI and LibreChat are separate projects with their own licenses, and some features or editions carry their own commercial terms. We check the current license and edition before we recommend one, and name any paid edition in the scope.
- Gateway fees
- Requesty plans and usage are set and billed by Requesty to an account you hold, unless the written scope says otherwise.
- Model usage
- Model usage is billed either by the gateway or by the provider, depending on how it is configured, to accounts you hold. It is not in our fees unless the scope says so.
- $99 consultation
- An optional, paid, 1-hour live session. Priced separately from setup, hosting, and support.
Full detail on how pricing works.
Proof
Related work
We have not published a case study of this exact stack, and we do not claim client results for it. The items below are related AI and automation work.
Automation showcase
Public examples of workflows, integrations, and tooling we have built.
View showcase →Case results are client-reported from our published case studies. We have not had them independently verified, and your results will depend on your starting point, data, and team.
Questions
Where these claims come from
Vendor terms change. These are the pages we start from, and we confirm the current version before scoping anything that depends on it. The DPA text itself is provided by Requesty on request.
- Requesty: security
Its stated security practices, including the DPA and EU gateway. It describes a SOC 2 Type II program as in progress, not certified.
- Requesty: privacy
Its stated logging, retention, and the scope of the EU endpoint.
- Requesty docs: EU routing
What the EU endpoint covers, and that model inference location is a separate matter.
- Requesty docs: compliance report
What its DPA covers and how to request it.
- Requesty docs: Open WebUI
How Open WebUI connects to the gateway.
- Requesty docs: LibreChat
How LibreChat connects to the gateway.
- Open WebUI docs: OpenAI-compatible providers
Connecting any provider that exposes an OpenAI-compatible URL and key.
- LibreChat docs: librechat.yaml
Custom endpoints with a base URL, API key, and model list.
Related
AI consulting & governance
Compare Claude, OpenAI, and a private stack before you commit.
OpenAI implementation
ChatGPT workspace setup or OpenAI API applications.
Claude implementation
A Claude-specific rollout: access, connectors, playbooks, and training.
Custom software development
When an interface and a gateway are not enough, scope a standalone build.
Tell us how you want AI set up for your company
Start a business wish list. If EU data location or a DPA matters, say so. The paid $99 consultation is a separate option if you would rather talk it through live.