AI CTRL User Guide
A practical, base-level enablement guide to the AI CTRL multi-model gateway for everyday users.
What AI CTRL Is
Expedient AI CTRL is a multi-model AI gateway built with a security-first approach. Instead of being locked into a single provider, your organization gains access to models from four major providers (OpenAI, Anthropic, Google, and Perplexity) through one secure interface.
A few principles are worth understanding before you begin, because they shape how you will work day to day:
- No profile building. AI CTRL does not learn your habits or build a profile on you the way some consumer tools do. Every new chat starts with a clean slate, and you provide the context each time.
- Direct model access. Your prompts go directly to the underlying models rather than through consumer apps such as ChatGPT or Claude.ai. This is why you supply context explicitly instead of relying on the app to remember.
- Single-tenant deployment. Your organization runs in its own dedicated, private environment. Your data is isolated and is not used to train models.
- Vetted rollout. New models are tested for stability and cost fit for a few weeks before being added, even after they become publicly available.
AI CTRL evolves on a roughly two-week release cadence, so the available models and some minor features may change over time.
Getting Started
Logging In
Use the single sign-on (SSO) button on the login page to authenticate through your organization's identity provider. If you can authenticate there, you can get in: there is no separate AI CTRL password to manage.
Mobile Access
AI CTRL is a Progressive Web App (PWA), so there is nothing to install from an app store. On your phone, visit the web address and add a home-screen shortcut. It behaves like a native app but is simply a secure redirect to your organization's gateway, still wrapped in your SSO authentication. Voice mode, described later, is especially convenient on mobile.
Working with Models
Choosing a Model
Open the model drop-down at the top of the screen to browse. You can scroll through the list or start typing to filter it by provider or model name. Each model includes a short description of what it is best suited for, so you can match the tool to the task. The models available to you are set by your administrators, so your list may differ from a colleague's.
Pinning and Defaults
Pin the models you use most to your sidebar for one-click access, and set a personal default so every new chat opens with your preferred model.
Not Sure Which to Pick? Use the Smart Model Router
If you are unsure which model fits a task, the Smart Model Router reads your prompt, checks which models you have access to, and routes the request to the best fit automatically. It runs the moment you submit.
Comparing Models Side by Side
You can send the same prompt to several models at once and view the responses in side-by-side "swim lanes." A practical limit is two to three models; beyond that, the screen becomes cluttered and hard to read.
Note on usage. Each model in a comparison is a separate request behind the scenes, so running several at once uses proportionally more of your resources than a single response.
Adding Context to a Chat
Because each chat starts fresh, you build the context yourself. The main ways to do that:
- File or web-page attachment. Attach a document, or point the chat at a specific web page. A web-page attachment scrapes only the page you link: it does not crawl linked sub-pages.
- Notes. A personal, text-based notepad you can reference inside a chat.
- Knowledge collections. Curated reference libraries, managed in the Workspace, for single-upload, multi-use material.
- Previous chats. Point to an earlier conversation so its content comes into the current one.
Built-in tools are invoked contextually: you ask for them in your prompt.
- Date and time awareness. Models do not inherently know today's date. When your prompt is time-sensitive, AI CTRL can pull the live timestamp.
- Chat search. Ask the system to search your past chats and it will run a retrieval search across your history to surface relevant conversations. The same works against your knowledge files.
Web Search and Image Generation
These two features are off by default and must be turned on per chat from the integrations menu. When enabled, their icons are highlighted so you know they are active. Simply typing "search the web" in your prompt will not trigger a web search: the toggle must be on.
- Web search routes through Perplexity, gathers citations, and passes the results to your selected model.
- Image generation and editing uses gpt-image-2 to create or edit an image, then hands the result to your chosen model.
Your administrators may restrict either feature for certain users.
Structured Data Tool (Excel / CSV)
Models often struggle with spreadsheets: long files get split into chunks, and headers and data types are lost along the way. The Structured Data Tool solves this by turning your file into a temporary, database-like table and letting you query it in natural language, which is translated into SQL-like commands behind the scenes.
Recommendation. Turn this tool on before uploading an Excel or CSV file you want analyzed, so you do not lose context.
Working with Responses
Once you have a response, you have several one-click options instead of typing a follow-up:
- Edit in place. Click into a response and adjust it directly (for example, to remove an em dash or an emoji) without asking the model to redo everything.
- Regenerate, More Concise, and Add Detail. Re-run, trim, or expand a response with a single button. Regenerated responses are saved as versions, such as "2 of 2," so you can page between them and delete the ones you do not want. This is a good way to keep your context clean.
- Switch models mid-conversation. Change the model at any point and keep going.
Exporting
Rather than asking the model for a download link, use the action buttons at the end of a response:
- PDF and Word export the full response. Your prompt is not included: only what the model produced.
- Excel exports table data only. If a response mixes narrative text and a table, the Excel export captures just the table.
- Anything else can be copied and pasted.
A PowerPoint export also exists but requires set up by an administrator before it appears.
Voice
AI CTRL includes text-to-speech (responses read aloud), dictation (speak instead of type), and a fully hands-free voice mode. These use whatever microphone and speakers your device has connected.
Organizing Your Work
Folders (Personal Projects)
Folders group related chats. At their simplest, they are directories that keep your sidebar tidy. Optionally, a folder can carry a baseline system prompt and attached reference files that apply automatically to every chat inside it, which is useful when a project will spawn several chats that all need the same context. Folders are personal rather than shared, and they can be nested; a nested chat inherits its immediate parent folder's context.
Managing Individual Chats
| Action | What it does |
|---|---|
| Pin | Keeps an important chat at the top of your list. |
| Share | Creates a read-only, point-in-time link another authenticated user can view. It is stateful: continuing the original chat will not update an existing share link; you would create a new one. |
| Clone | Lets someone who received a shared chat make their own fully interactive copy to continue it. Cloning does not affect the original. |
| Unshare | Revokes a share link; anyone visiting the old link is redirected to the base gateway. Useful for putting a self-imposed time limit on sensitive content. |
| Archive | Hides a chat from your sidebar without deleting it. Retrieve it anytime from your profile. |
| Delete | Permanently removes the front-end chat record. Back-end compliance logging retains a record regardless; deletion only affects what you see. |
Personal Settings
Personal System Prompt: Your Highest-Impact Setting
By default, this is empty, which means the model guesses your tone, formality, and technical level every time. Filling it in puts your interactions on rails and reduces the back-and-forth. A good starting set of contents:
- Who you are: your name, role, and organization.
- The products, services, or workflows relevant to your daily work.
- Communication preferences: tone, formatting, jargon tolerance, and any do's and don'ts.
Tip. A great first task for any new user is to pick a model and ask it to "help me build my personal system prompt," then brain-dump the relevant details. Start small, see what works, and refine over time. Once saved, this context applies to every chat.
Interface Settings
Most interface settings are aesthetic, such as widescreen mode and chat-bubble style, but two are functional:
| Setting | Purpose |
|---|---|
| Allow user location (off by default) | AI CTRL does not read your location unless you enable this. Useful if you travel often and location matters to your chats; otherwise, you can simply state your location in a prompt when relevant. |
| Copy formatted text (on by default) | Preserves Markdown formatting (headings, bold, lists) when you copy or export. Turn it off if you want flat text. |
Memory (Beta)
Memory is similar to the feature in consumer tools but stricter. It is 100% manually managed (the system never auto-creates or auto-deletes memories) and it requires you to opt in. Add short, durable pieces of context (the classic example is "I am learning Spanish") that the model can draw on when relevant, and remove them when they no longer apply. The manual-only design prevents accidental capture of sensitive information. Your administrators can disable this feature globally.
Data Controls
- Export and import your chats as a JSON file for backup.
- Shared chats list: review and unshare anything you have shared.
- Archive all: hide every chat at once, handy after test sessions.
- Delete all: permanent and unrecoverable for the front-end record; double confirmation is required.
File Manager
Review, view, or delete any file you have uploaded. Deleting a file removes it from chat context so it can no longer be referenced. You see only your own uploads.
Account Password
This field only applies to direct-access (non-SSO) accounts. If you sign in through your organization's SSO, changing a password here does nothing: manage your credentials through your identity provider. If you have any questions or concerns, please contact your system administrator.
What's Coming
This is a base-level guide. Further sessions cover the Workspace (custom models, knowledge collections, prompt templates, and skills) and compliance. End-user training typically includes a 101 session (a basic overview) and a 201 session (workshopping custom workflows), run with your assigned client success manager.
Frequently Asked Questions
How does model licensing/access work behind the scenes?
Expedient holds enterprise agreements with each model provider. When a new model is released to the provider's public API, Expedient runs it through several weeks of stability and cost testing before releasing it to clients.
If I use the Smart Model Router, how do I know which model handled my request?
The router evaluates your prompt and the models available to you, then picks the best fit at the moment you submit, automatically, per message.
Do folders work like directories?
Yes, and they can also carry their own system prompt and attached files that apply to every chat inside.
Does memory pop up automatically?
No. You opt in, and you add each memory entry manually. Nothing is automatic.
What happens to a deleted or shared chat on the back end?
Every interaction through the gateway is logged in a separate compliance platform, regardless of what you delete on the front end.
Why did a long "deep research" query take several minutes?
Deep-research models can take several minutes. A background process keeps the chat alive while the model works.