Every capability of Lima Juno — Agentic AI has its own page: what it does, the steps it runs, what you see on screen, and the evidence behind each number.
The order matters: each stage produces what the next one needs, and trust grows in the same order.
Each stage links to its first page. The pill row at the top of every capability page carries the rest of that set.
Four pages on connecting a source, choosing tables, reading in place and keeping up with changes.
Pick a connector, give the connection, choose the tables. The rows stay in the source system.
Over forty kinds of source: ERP, planning, warehouse, service, lakes, databases, files and streams.
One question can read three systems at once through the federated query engine, with your access applied.
A scan lists what is new, changed or gone in a source, and you choose what to adopt.
Five pages on the ontology, the proposals Juno makes, the review screen and linking the same thing across systems.
Juno proposes which column carries which ontology attribute, with a confidence and a reason for each.
Your team confirms, adjusts and completes the mapping on one screen, and every change is recorded.
What the ontology holds: objects, attributes, relationships and measures, and how you add your own.
An index of your tables, the values in each column, and the words your business uses for them.
The same supplier in three systems is recognised as one, by matching the values across columns.
Eight pages on ChatLima: from a typed question to a checked answer, in the places where people already work.
Ask your supply chain a question in your own words; the answer arrives with its working on show.
Five steps turn a question into the right query, over one source or across several.
A table, a chart or a short narrative, presented the way you meant the question.
A follow-up keeps the thread: the period, the plant and the business objects carry over.
The same question gives each person the answer their role is cleared to see.
Every answer shows the terms it used, the systems it read and the query it ran.
Root causes with their supporting data, comparisons with history, and short-term forecasts.
The ChatLima screen, Microsoft Teams and Slack, SmartApps, an API and a Power BI connector.
Five pages on teaching Juno your vocabulary, keeping its grounding current and measuring its accuracy.
Vocabulary files, acronyms, abbreviations and aliases teach Juno the words your company uses.
Mark an answer right, correct a term, save a verified pair: the next answer improves.
Named policies, named plans, what Juno knows about your estate, and the data it was trained on.
Save a line of questions as a walkthrough for a recurring review, and share it with the team.
Score fifty of your own questions for correctness, speed and access rules, after every change.
Nine pages on the task agents, the plan behind them, the rules engine, approvals and confirmed write-back.
Ten task agent roles, from PO cancellation to allocation: one runs in production, nine are designed, yet to be built.
State the intent in plain language; review, edit and simulate the plan Juno drafts.
The Lima Rules Engine executes a published plan the same way on every run.
State what to watch in plain language; a schedule, an interval or an event starts the check.
Each recommendation carries its basis, its confidence and the value at stake.
At the supervised level a named person approves each action, with clocks and an escalation route.
Approved actions go through the interfaces your ERP supports, and the ERP confirms each one.
When two agents want opposite changes to one field, weights decide inside a band; beyond it, a person.
An action can be an event, a workflow, an API call, a database write or a file.
Eight pages on the learning loop, the audit trail, versions, cost control and the manager’s view.
Every accept, edit and override is recorded and used, so the next cycle is better.
Juno checks what happened against the objective the agent was given, inside a set window.
One record per action: signal, rule version, limits checked, approver, write-back and outcome.
Every change is a new version, and simulation on past data comes before release.
Budgets per agent and per team, a meter on every model call, and the routing choice.
Choose the cloud model for design time; a small model inside your estate serves run time.
Twelve layers of security, single sign-on, role-based access, and the asking person’s own rights.
Run traces for each decision, and the manager report: corrections, overrides and exceptions.
Six starting points; each one takes about fifteen minutes to read.
Figures carry the mechanism; the prose carries the steps; the labels carry the evidence.
Each capability page opens with a figure of the mechanism, then a short band of figures with their evidence labels. A “step by step” section follows the work in order, and the sections after it show the screens, the tables and the interfaces behind it.
The running example runs through the whole set: one planner, 1,240 exception messages on a Monday, and a cancellation window that closes on Thursday. Each page carries the next instalment and a link to the page that follows.
Every page closes with the terms it used, the questions buyers ask, and internal review notes that say where each claim comes from and who still has to confirm it.
Read automatic mapping, show the working, the rules engine, approvals and the audit trail. Those five cover meaning, grounding, execution, oversight and evidence.
Each stage is useful on its own. Registration and mapping give ChatLima; the act and learn stages build on the same ontology and access rules when you are ready.
Every figure carries one of five labels: production, anonymised engagement, showcase estate, estimate or dated source. The label sits beside the number.
Each one builds on the one before it.
Reads your rows in place.
Connect and move ↗One record per real thing.
Master and govern ↗Apps for every decision.
Operate and decide ↗Asks and acts, at your level.
Ask and act ↗Bring two systems and one decision your planners make every week. In the first week we register the data, confirm the mapping with your team, answer your questions in ChatLima and run the first agent at the level you choose.