Lima Juno reads your supply chain data where it lives and maps it to a built-in supply chain ontology your team confirms.
One planner’s Monday and one agent, from the question to the confirmed action. Each chapter builds on the two before it and closes on the question the next one answers.
One production fact, one showcase Monday, and a map of designed roles: each carries its own label.
The PO cancellation agent runs at a high-tech manufacturer. Planners author the cancellation policy in plain language, the Lima Rules Engine executes it on federated ERP data, and write-back to the ERP waits for a named person’s approval. Every action it has taken traces to a versioned rule, and run time costs zero tokens, because the rules engine decides and a model inside the estate explains.
It is the one agent on this site that runs today. Lima’s agent blueprint is a design of 960 agent roles across five end-to-end chains. It is the map of what Juno can be configured to do, and the first agents are built from it one at a time, the way the PO cancellation agent was. Chapter 8 shows the map; chapter 5 shows the one that has been built.
The four rows on each side are the ones chapters 1 and 5 carry. Production facts are labelled production; the Monday is the showcase estate.
How much an agent may do is your decision, agent by agent, and it changes on evidence.
All five are roles on the blueprint. The A2 role is the one in production; the four marked designed are yet to be built. The blueprint designs PO change and cancellation at A3; the showcase estate started it at A2 and raised it on evidence. Chapter 3 tells each scene in full.
The ladder turns “how much AI” from an open-ended technology question into a per-agent governance setting you control. A level goes up on evidence: recorded cases pass a test, a named owner approves, an escalation route is live. It comes down in one step, at once.
Six stages that build on each other, and the first week walks them one a day.
Each stage produces something the next one needs. Registration gives Juno the tables. Mapping gives those tables one meaning in the supply chain ontology, Lima’s built-in map of supply chain things and how they relate. With one meaning, a question in plain language becomes the right query, and a policy in plain language becomes the right rule.
The order also matches how trust grows. People first see their own data answered correctly, then teach it their words. Then they let an agent draft actions for approval, and later delegate the routine cases and keep the exceptions.
ChatLima is one of Juno’s feature sets: the conversation, grounded in the same ontology the agents use.
38 open lines worth $412,000 are inside their cancellation window, which closes on Thursday 24 September. The PO cancellation agent ran at 06:00 on rule version 3: 29 drafts in D. Okafor’s queue, and 9 lines with him as buyer, each carrying its reason.
| Lines | Count | Open value | Where they are |
|---|---|---|---|
| Under $25,000 | 29 | $187,300 | D. Okafor’s approval queue, as drafts |
| $25,000 and above | 9 | $224,700 | D. Okafor as buyer, for his judgement, with the reason for each |
| All | 38 | $412,000 | window closes Thursday 24 September |
SELECT CASE WHEN l.open_value_usd < 25000 THEN 'under' ELSE 'at or above' END AS band,
COUNT(*) AS lines, SUM(l.open_value_usd) AS open_value
FROM ontology.purchase_order_line l
WHERE l.status = 'OPEN'
AND l.cancel_window_end >= CURRENT_DATE
AND l.projected_excess_weeks > 8
GROUP BY bandThe story first; the use cases, ChatLima, the reference pages and the analyst guide after it.
Five chains, the twenty use cases the market is deploying now, and the map of everything else.
Open the use cases →The conversation on its own page, and the two-week bake-off on fifty of your own questions.
Open ChatLima →One page per mechanism, from data registration to the token manager, for depth after the story.
Open the reference →How analysts categorise Juno, with each required feature mapped to the page that shows it.
Open the guide →One in production: the PO cancellation agent, at a high-tech manufacturer. The 960 roles of the blueprint are designed and yet to be built; each is configured from the same parts, the way that agent was.
The rows stay in your systems and Juno reads them there, so work starts the day the systems are registered. Where history at scale is needed, Quanta lands a governed copy in your lake and Juno reads it there.
You do, agent by agent, with the level setting from A0 to A4. The showcase estate started its agent at A2 and moved lines under $25,000 to A3 after six weeks and 212 approvals.
At design time, the cloud model you choose: Gemini, Claude, OpenAI, Llama or another. At run time, a small language model inside your estate, at zero token cost. The token manager meters every call.
The matrix maps the market’s own criteria to what Juno provides today, with the evidence label the site uses everywhere. The pages beneath are the ones an analyst reads first.
| What evaluators score | What Juno provides | Evidence | Where it is shown |
|---|---|---|---|
| Decision modelling | A policy stated in plain language, compiled into rule rows anyone can read | In production | The page → |
| Decision composition and execution | The rules engine runs a published version the same way every run | In production | The page → |
| Collaboration between people and agents | Four patterns, four screens, nine configurable settings per agent | Showcase estate | The page → |
| Autonomy and control | A0 to A4 set per agent; promotion on evidence, demotion in one step | Showcase estate | The page → |
| Monitoring and observability | Run traces per decision, manager reports, token budgets per agent and team | Showcase estate | The page → |
| Governance and audit | One record per action: signal, rule version, approver, write-back confirmation | In production | The page → |
| Natural-language analytics | ChatLima answers on data in place, with the working shown and the access policy applied | Showcase estate | The page → |
| Learning from outcomes | Every accept, edit and override recorded; outcomes checked against the objective | Showcase estate | The page → |
The analyst guide for Juno → · Trust center → · Request a briefing →
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 ↗Nine chapters, one planner’s week and one agent followed from a sentence to production. Start with the Monday problem, or ask for the governance demo on the showcase estate: the approval queue, the level review, the audit record and the manager report.