
How to Build Your AI Command Center with Bento
Stop juggling scattered AI tools. Build a centralized AI command center in minutes — one place for your integrations, knowledge base, automations, and custom apps.

Quick Summary
Updated for AI discoveryBento gives every workspace a private cloud VM and persistent AI agents. To set it up: (1) start one durable session with a real outcome, (2) organize workspace files like you would brief a new hire, (3) add stable facts and sources to Train, (4) keep scheduled results in session history and files, (5) connect tools through Integrations, and (6) ship internal or external apps from the same workspace. Specific context matters more than novelty.
Questions this page answers
This guide is the exact walkthrough David, Bento's founder, gives to new customers on their first call. It is not a feature tour. It is the order of operations that turns a new workspace into working AI agents — what to do first, what to skip, and the mindset shifts that separate the people who get 10x value from the people who treat Bento like a fancier ChatGPT.
If you just signed up, read this once before you do anything else. If you're an implementation partner setting up Bento for a client, this is the playbook.

Start with one real session
02Garden your knowledge base
03Train reusable context
04Schedule recurring work
05Connect any tool with an API
06Ship internal and external apps
07Security and your own server
08Billing without surprises
09Pick the right model
10Connect external services
11Workspace identity and app domains
12The mindset shift
13FAQ
Sign in and Bento opens on Home, where you pick an agent and start a conversation. Each conversation is a durable session that keeps its history. Start with one real outcome instead of trying to configure the entire workspace at once.
The biggest predictor of a useful first result is the quality of the brief: the desired outcome, the source material, the constraints, and the test that tells the agent it is done.
Think of it like this. If you transplanted every memory you have into another person's brain, that person would behave a lot like you — same opinions, same instincts, same shortcuts. The model is the brain. Your memory is what makes Bento act like your agent instead of a generic assistant.
What to do on day one
Start a session with one real job. Upload the sales deck, strategy document, ICP notes, style guide, or other files the job depends on, then state what a finished result must contain.
When the session asks a question, answer it directly. The answer becomes part of that durable work history instead of disappearing after a one-off prompt.
Bento builds memory implicitly through conversation and explicitly through Train. You're not retraining the model. You're giving it context.
The Drive in your workspace is a real file system on a real cloud server. You can upload PDFs, drag in files, or create folders, and Bento can read everything in there. To turn a public web page into markdown, add its URL in Train (step 3) or ask the agent to fetch it.
The way to think about it: set it up the way you would organize a Google Drive for a brand new employee.
/files
/company
one-pager.pdf
pitch-deck.pdf
style-guide.md
/customers
target-icp.md
case-studies/
/products
pricing.md
roadmap.md
/sales
discovery-questions.md
objection-handling.md
This is what David calls "gardening." You don't have to do it constantly. A quick sanity check every couple of weeks is enough — move anything that obviously belongs somewhere else, prune anything stale.
A well-organized agent instructions file links to sub-documents, which link to other files. The agent navigates your file system the same way you do. Better structure = smarter agent.
Power-user tip
The Files API exposes the same workspace file tree as the web app, so automated workflows can create, read, move, and publish artifacts without a second storage model.
Train is where a workspace keeps stable facts and sources that should be reusable across sessions. Open it from an agent's conversation header, add files or public page URLs, and start a training job. The resulting memories appear in the memory library, where you can inspect and correct them. Training at the workspace level gives every agent the same memory.
After a strong session — for example, an SEO audit with the right internal-link, cannibalization, AEO, and ranking checks — extract the durable facts and standards that future work should reuse.
"Save the stable SEO criteria and our company-specific priorities so later sessions can reuse them."
Keep those standards in Train and the supporting files in the workspace. Future sessions can retrieve the same business context without inventing it again.
This is the part that does not exist in a disposable prompt window. Bento keeps the session history, trained context, and files together so useful work compounds.
What to add to Train
Anything that has the shape: "domain expert + repeatable workflow + tedious to do every time." Examples: weekly competitor monitoring, SEO audit per page, customer onboarding email drafting, inbound lead enrichment, monthly board update prep. If a fact or standard should survive this session, add it to Train.
Reusable context should be specific to your company: its keywords, priorities, terminology, customers, and trusted data sources. That specificity is the point.
Pruébalo en Bento.
Posts, reels, anuncios, videos y apps en un link, hechos con la mejor AI. Una sola suscripción.
Nada que configurar
Bento's workspace VM stays available when your laptop is closed, so the agent can run scheduled jobs and keep their results in durable sessions and files.
Here is what it looks like in practice. Start a session and give the AI an instruction:
"Every hour, check our connected support inbox. Every day at 9am, update the growth report. Every Monday, run the weekly SEO check."
Bento sets up the scheduled work on the workspace VM. Its session events and output files leave an inspectable record of what ran and what changed.
The big win over hidden cron jobs:
Moving existing cron jobs
Start a session and ask Bento to inventory the cron jobs on the workspace VM, explain what each one does, and migrate only the jobs you approve. Keep the resulting runbook in workspace files.
Bento ships with thousands of ready-made integrations — Google Calendar, Slack, Notion, Linear, Stripe, Gmail, Telegram, you name it. Browse them under Integrations in the sidebar and connect with one click.

What most people miss: Bento can connect to anything with an API, not just the ones in the catalog.
The way it works: Bento is, at its core, a coding agent. Every AI agent is. Even when you ask it to do "research" or "sales prospecting," it's actually writing and running code under the hood. So if you want to connect a random fifteen-year-old billing system that nobody else integrates with, you just ask:
"Does Sentry have an API? Connect to it. Here's our API key."
Bento looks up the docs, writes the integration, and calls the API through the credential boundary you authorize. For catalog integrations, credentials remain with the server-side provider connection rather than being returned to browser JavaScript.
Secret handling
Prefer the Integrations screen for supported services. Bento keeps provider credentials on a server-side connection boundary, scopes workspace credentials, and does not expose raw keys to browser JavaScript or session events.
Because Bento has its own server and runs a coding agent, it can build apps. Real ones.

There are two flavors and you'll probably use both:
Internal apps. A dashboard, an admin tool, an "internal use cases" page that links to your best workflows. Bento builds them, hosts them on your server, and you can share the URL with your team. We use this constantly — every cool tab in our own Bento workspace ends up linked from one internal app we maintain.
External apps. A landing page, a marketing site, a customer portal. Same flow. You ask, Bento ships, the app gets its own Bento address by default (<app>--<workspace>.edge.bento.ing), and you can wire your own custom domain (like consigliere.com) by setting up the DNS record. We currently host customer-facing marketing sites for several companies entirely inside their Bento workspace.
Hosting, custom domains, SSL — all included. No "deploy" button to find. The app is just there once Bento writes it.
Real example
A bartender in Texas who runs a seven-year-old bar was paying a dev agency $3,000/month for his website. He moved his entire site into Bento and now pays a fraction of that for a better one that he can iterate on himself, in plain English.
Set up your first agent
One workspace. Agents. Drive. Train. Integrations. Apps. Your private cloud server.
Nada que configurar
The single most important thing to understand about Bento's security model: every workspace is its own server.
Not a shared agent process. Each workspace receives a dedicated single-tenant virtual machine. Its files and processes are isolated from other customer workspaces.
This isn't just a marketing label; it is an architectural boundary. A dedicated computer lets Bento write code, install tools, retain files, and continue work across sessions without sharing a workspace runtime with another customer.

Practically:
If you're at a company with compliance requirements, this single-tenant model is usually easier to defend than a typical multi-tenant agent platform, not harder.
Bento has three monthly plans: Starter, Pro, and Max. Two things to know.
Monthly credits reset. Each plan includes a monthly credit bucket. It resets at renewal and unused monthly credits do not roll over, so choose the plan around the work you expect Bento to own in a normal month.
Overage uses your paid reserve. When monthly credits run out, Bento draws from the paid reserve. The designated payer can refill it manually or enable auto topups; every subscription charge and topup produces an invoice.
How to avoid sticker shock. Three settings, all under Top-ups and limits on the designated payer's billing page:
| Setting | What it does | Recommended |
|---|---|---|
| Monthly spend limit | Hard ceiling per month. Bento stops at this number. | Set it to your real budget. |
| Top-up amount | How much credit gets added when you run low. | $20 minimum for casual use; higher if you use Bento hard. |
| Top up below | The balance at which auto top-up kicks in. | Don't set this too low — usage comes in bursts of $5–$50 at a time, and your credit card processor will start blocking too-frequent charges. |
Workspace settings → Usage charts spend by day, week or month and lists every credit grant, top-up and usage charge, newest first. Each turn in a session also shows its token usage per model. You can always see what cost what.
Bento routes each task to the strongest model for it, across providers, while preserving one durable session history. As new models ship, the routing moves with them. You choose a quality level rather than a model: Max routes to frontier models for the strongest results, Standard balances quality, speed and cost, and Economy uses low-cost models for scheduled maintenance work. Workspace owners set the default quality in workspace settings, and Advanced lets you pin a specific model for a session.
Default to Max when you're getting started. Especially in the setup phase. Stronger models are better at instruction following and deciding whether a workflow is worth saving as a reusable playbook. The extra cost can be worthwhile while the workspace is taking shape.
Once Train context and reusable playbook files are tuned, Standard or Economy can handle workloads that do not need the most expensive reasoning — Q&A over your knowledge base, drafting a quick reply, or summarizing a report.
Prompt rewriting trick
Want to run a workflow on a cheaper quality level but the output isn't as good? Run the workflow on Max first. Then run the same prompt on Standard or Economy and ask: "Keep improving your output until it matches what the Max run produced." The cheaper models will iterate, and along the way they'll write a longer, more explicit version of the prompt that you can reuse. Cheap models often just need more guidance — let AI write the better prompt for you.
If a workflow depends on one model's particular style, pin that model with Advanced. Otherwise let the routing pick, so the workflow keeps improving as better models ship.
Bento connects external systems through the Integrations screen. The agent can then use authorized actions inside a session without exposing provider credentials to the browser.
Email. Connect a supported email provider and give the agent a bounded task. Common patterns:
The resulting tool calls and outputs remain attached to the session that initiated them. Disconnecting the integration revokes future Bento access.
Bento keeps workspace identity and published apps separate. You can:
The only thing we don't currently support fully white-labeling is the root Bento app URL. The pragmatic workaround: build your customer-facing experience as an app inside Bento, point your own domain at it, and the customer never sees the word "Bento."
Implementation partners
If you're an agency or partner setting up Bento for clients: the cleanest pattern is to create a workspace on behalf of the client, configure it, then invite them as an Owner, make them the designated payer, and leave the workspace yourself when you're done. The workspace is no longer tied to your account once you leave. Bento won't let the payer or the last owner be removed, so hand both over to the client first.

If you take one thing away from this guide, take this:
Using AI well is a little like gambling.
You're betting some tokens — call it $1, or $10, or $100 — that the AI will produce something that returns multiples of that value. Sometimes it doesn't. Sometimes the run goes sideways and you've spent $20 on nothing. But over a month, if you know roughly what you want, the expected value is heavily positive. Our top users spend hundreds of dollars in tokens a month and get tens of thousands of dollars of work back.
This mindset only works if you know what you want. Vague prompts on expensive models with no context in place will burn money fast. Specific outcomes, backed by Train sources and reusable instructions, are easier to evaluate and repeat.
The corollary: don't try to learn AI by exploring randomly. Pick one boring, valuable, repeatable workflow in your business — answering customer support emails, doing weekly SEO audits, updating your website, drafting outbound sequences, building an internal dashboard — and make Bento good at that one thing. Then add the next. People who try to "use AI for everything" on day one usually use it for nothing by week four.
Pruébalo en Bento.
Posts, reels, anuncios, videos y apps en un link, hechos con la mejor AI. Una sola suscripción.
Nada que configurar
Set up your workspace
One private server. Sessions, files, trained memory, integrations, and apps.
Nada que configurar
Posts, reels, anuncios, apps y juegos, hechos con la mejor AI.
Créditos gratis para probarlo todo. Nada que configurar.

Stop juggling scattered AI tools. Build a centralized AI command center in minutes — one place for your integrations, knowledge base, automations, and custom apps.

Bento AI makes Founder Mode an operating system. Run a 3-person company like a team of 50. Ship faster. Skip middle management. Stay close to the product.

MCP vs skills, tools, and CLAUDE.md: see what each does, how they work together, and how to troubleshoot a broken agent workflow.

Speed up client emails and renewal follow-ups with an AI drafting system that keeps communication consistent.

Reduce carrier portal rekeying with AI that extracts ACORD data and powers automated carrier submissions across portals.

AI transformation is 9 parts process, 1 part AI. The 10-step playbook for making any workflow AI-native — and where most teams quietly get stuck.