G5 reads your operation, simulates the options and recommends with evidence.
Your people decide, and your systems execute. G5's agent loops help planners, engineers and managers solve production problems on your own data, with every number traced to its source and every forecast shown with the error it measured.
Decisions with their evidence
Every recommendation comes with its options compared, its numbers traced to their source and its forecasts' measured error — ready for the person who signs.
You stay in charge
G5 reads what you allow and recommends. A named person approves every critical change, and every action is logged.
Start with one loop
Start with the loop for your biggest problem — SR, APS, Twin or QA — on the exports you already produce. Add the next when you're ready: none requires another.
Four decision loops over one data loop
Each decision loop works toward one production goal: it reads its own data and ends in its own decisions. Under them, the data loop keeps that data ready and reliable. They share one foundation — the lakehouse with your data checked against a declared standard, the planning engine, machine learning that reports its own error, and approvals with named owners — and none requires another. A plant starts with the loop for its biggest problem.
What should we make, and when?
Is my production running as expected?
Is my process running as expected?
Are my quality results as expected?
Every decision loop reads from the same lakehouse. The data loops keep it ready and reliable — and, where regulation asks, turn it into validation evidence.
Is my data ready and reliable?
How do I validate G5 in a regulated plant?
How a G5 loop works
From your goal to a decision, in seven steps every decision loop follows. Pick one:
How DataBot/SR plans your production
Seven steps, one example from an injection-molding plant.
Perception
Takes in requests, files, system data and sensor signals.
Validation
Checks the data before trusting it.
Reasoning
Works out what is possible, and why.
Planning
Turns the goal into options and compares them.
Reflection
Checks its own work before you see it.
Action
Uses tools to get the work done.
Memory & learning
Remembers what happened and improves.
You stay in charge
The agent acts on its own within limits you set; critical changes wait for a named approver, and every action is logged.
Is my production running as expected?
How DataBot/APS watches the production flow and re-plans when it drifts.
Perception
Watches the production flow as it runs.
Validation
Tells a plant problem from a data problem.
Reasoning
Finds what drifted from the plan, and why.
Re-planning
Re-plans around the problem.
Reflection
Checks its own work before you see it.
Action
Uses tools to get the work done.
Memory & learning
Remembers what happened and improves.
You stay in charge
The agent acts on its own within limits you set; critical changes wait for a named approver, and every action is logged.
Is my process running as expected?
How DataBot/Twin watches the process parameters behind quality and proposes re-setups.
Model validation
Before activation: proves the model on your data.
Perception
Watches the process signals behind quality.
Validation
Checks each signal before trusting it.
Reasoning
Spots drift before a limit is crossed.
Re-planning
Proposes a re-setup and re-plans around it.
Reflection
Checks its own work before you see it.
Action
Uses tools to get the work done.
Memory & learning
Remembers what happened and improves.
You stay in charge
The agent acts on its own within limits you set; critical changes wait for a named approver, and every action is logged.
Are my quality results as expected?
How DataBot/QA watches quality results, proposes re-setups and re-plans.
Model validation
Before activation: proves the model on your data.
Perception
Watches quality results as they come in.
Validation
Checks each result before it counts.
Reasoning
Spots drift while results are still in spec.
Re-planning
Proposes a re-setup and re-plans around it.
Reflection
Checks its own work before you see it.
Action
Uses tools to get the work done.
Memory & learning
Remembers what happened and improves.
You stay in charge
The agent acts on its own within limits you set; critical changes wait for a named approver, and every action is logged.
The plan, solved against the capacity you really have
Every G5 loop plans on the same engine. DataBot/SR runs your monthly supply review and S&OP on it, and the other loops use it to weigh their options against capacity: routings with their candidate machines, shift calendars, efficiency and downtime, solved at finite capacity in about a minute and explained by the agent in your language.
Service level, stated honestly
Served versus asked, gross and net side by side, and every shortfall tied to the machine or line that constrained it.
Stock and priorities by rule
Stock buffers by days of cover, priority groups with their floors and ceilings, and strategic targets in the same plan, each with the capacity it costs.
Options compared
Fork a scenario, change one thing — a shift, an efficiency, a buffer — run it again and compare it with its parent, right in the conversation.
The Control Tower
The central screen: the plan's service level, its constraints and the agent's reading of the run — the same numbers for planners and managers, and every run stored with the data it read.
Models trained on your own history — with the error on the label
Every G5 loop that predicts something shows how good the prediction is. Models are trained on the data your operation already produces, and each one reports its error measured on data it did not learn from — so you know how far to trust a forecast before you plan on it.
Demand forecasting
Forecasts from your own sales and order history, by product, ready for the plan — with the error they actually had on past months, set beside a simple seasonal forecast.
Your problem, your data, a model
Bring a data file and say what you want to predict. The agent profiles the data, prepares it, trains and tells you the measured error — and names what is missing when the data can't support a model.
Signals over time
Sensor and process series become one prediction per case — which cycle, which batch — with the signals that matter named.
From your own exports to a decision
No integration project to begin. Bring the problem that costs you the most: our team guides each step, on the files you already export.
Talk to an expert
Tell us the biggest problem — the plan, the schedule, the process or quality. It picks the loop you start with.
Send your exports
The files that loop reads, as spreadsheets or extracts. Each one is checked against the G5 standard, and anything missing or malformed is refused by name.
A trial on your data
The loop works on your own exports with our team: the decision it brings, the error it measured and anything still missing, named.
Contract the loop
Per loop and per plant, in the edition that fits. From then on, what changes goes to its owner as an approval card, and the record keeps who, when and why.
Then, the next loop
Each loop starts when its data is ready, building on what the first one brought in. One plant or many: each plant or business unit gets its own database, with its own people and permissions.
Three editions, the same loops
Every edition runs the same loops on the same release. What changes is how your data is isolated, where it runs and how you contract it.
G5 Cloud
DataBot runs the service on shared infrastructure in a standard three-region cluster — Virginia, Ohio and Canada — with a database for each customer and the models on DataBot's GPUs. A plant starts on files it uploads.
G5 Enterprise-Cloud
The same standard three-region cluster, with a container and a database for each customer and the models on DataBot's GPUs. A plant starts on files it uploads.
G5 Enterprise-Private
A dedicated three-instance cluster in the regions DataBot supports — Virginia, Ohio, Canada, São Paulo, Mexico, Frankfurt, Ireland and Paris — with a negotiated data-processing agreement and service level, and an annual contract with the DataBot team.
G5 reads, recommends and waits for your yes
The line that matters runs between reading and control. G5 reads the data you allow, recommends with evidence and takes the decision to the person who owns it. What happens next in your plant stays under your control.
Read
The exports, files and documents you choose to share; plant data only through read-only connections.
Recommend
Options compared, numbers traced to their source, forecasts with their measured error.
Approve
A named person decides on an approval card. Rights live in one auditable grant store, and the server — not the AI — checks every action.
Execute
Your people and your systems carry out the decision. G5 never writes to your control systems.
Your data, our own AI — no third parties in between
G5 runs in DataBot's cloud on our own GPUs, with language models we host ourselves — not on third-party AI APIs. Your production, cost and quality data is processed on our infrastructure and never used to train public or outside models.
Our own AI
DataBot's GPUs and self-hosted models. No external AI APIs.
Your data stays yours
Never used to train public or third-party models.
Authority lives on the server
Every action is checked against your grants by the server — the AI is never the gatekeeper.
Isolated by design
Each customer has its own database; the Enterprise editions add their own containers or a dedicated cluster.
Five nines, measured — not promised
99.999% is the availability we measure, not a target on a slide. It comes from the architecture: G5 runs active in three regions — Virginia, Ohio and Canada — so losing a server, or even an entire region, doesn't take the service offline.
Three active regions
All three regions serve at the same time, and a global front door sends every request to a healthy one.
The service in every region
G5 runs in each region, and every instance registers itself as it starts. If one stops, the others keep serving while it restarts.
Three copies of your data
The lakehouse keeps every table in three replicas, one per region. Losing a region loses no data and doesn't stop the service.
Updates without downtime
New versions roll out one instance at a time, each verified before the next — the service stays up while it updates.
No admin console to learn. Just ask.
Managing G5 works the way using it does: in conversation. Invite a colleague, grant access, load a file, change a setting — the agent prepares it, and anything critical waits for the person responsible to approve it. And because G5 is a service, we run the platform for you: updates arrive continuously and never break what you already use.
Managed in conversation
People, access, data loads and settings, all through the agent. No forms, no tickets.
One place for every right
Each permission is a single, auditable grant with a clear role — responsible, accountable, support, consulted, informed (RASCI).
Approvals where they matter
Critical changes become approval cards. A named person decides, and the record keeps who, when and why.
We run it for you
Software as a service on DataBot's cloud: continuous, backward-compatible updates, monitored around the clock.
Plugs into the systems you already run
G5 doesn't ask you to replace your ERP or rewire the plant. It reads the extracts and files you already produce, validates every load against a declared standard — refusing bad data by name instead of loading it silently — and reads machine data over open industrial protocols.
ERP and business systems
Files and documents
Machines
Let's talk about your operation
Tell us about the production problem — a plan nobody trusts, orders that ship late, a line that never hits its numbers. We'll show you a G5 loop working on it, starting with your own exports.
Talk to an expertTechnology Park, Suite 1403
Av. Doutor Altino Bondesan, 500
São José dos Campos, SP — Brazil · 12247-016