Every capability below exists because a customer brought us the problem. We're not another system to manage, either. Junipix is the AI decision layer that sits on top of the systems you already run: ERP, QuickBase, QuickBooks, Excel, EHS, WMS, TMS, CMMS. We don't replace them. We connect them, interpret them, and turn them into answers your people can act on.
A note on counting. Depending on how you slice it, this page lists three agents, or twenty-one, or one platform with a lot of strong opinions. The AI industry hasn't agreed on what an "agent" even is, and we've heard every argument, mostly from each other. So we organized this the way the work is actually organized and numbered it so we can all point at the same thing. If your procurement form needs a category name, we'll match whatever it says.
Typical deployment: 6 to 8 weeks, $10,000 to $40,000 depending on scope, configured to your data and terminology. Some of this is deployed in production today, some is pilot-ready, and we'll tell you which is which when we talk.
Your policies lead, the regulations back them up, and your people get one controlling answer with both citations. OSHA and DOT are built out today; the same structure extends to FAA, PHMSA, FSMA, EPA and whatever agency your operation answers to next.
"What does OSHA actually require here?"
Plain-English answers with the exact citation, scope, and next steps.
Then it gets layered:
"The reg says one thing and our manual says another. Which controls?"
Reconciles OSHA against your internal standards and returns the stricter requirement, cited on both sides. One question, one answer.
"We're on their site. Whose rules apply?"
Contractors live under three rulebooks: OSHA, your policy, and the client's site requirements. This layers all three and returns the controlling answer before your crew clears the gate.
"Does everything we've written actually agree?"
Your manuals, procedures, and training docs cross-referenced against each other and the regulations. Conflicts and gaps get flagged for review instead of discovered in an audit.
Same structure, pointed at the road: driver qualifications, hours of service, vehicle standards, hazmat. Cited answers in plain English.
Then layered:
"DOT allows it. Do we?"
Your fleet and driver policies reconciled against the federal requirements, with the stricter standard returned as the answer your dispatcher acts on.
"The shipper has their own rules. Now what?"
Customer routing guides and shipper requirements layered onto DOT and your own policy, so one question covers all three before the load moves.
"Do our contracts, rate cons, and driver files line up?"
Transportation documents cross-linked and checked against each other, with the disagreements surfaced while they're still cheap to fix.
Dense filings where the difference between two provisions is real money. This family turns the filed record into commercial intelligence, answering only from authoritative sources with the exact provision cited.
"Is our tariff library actually current?"
Continuously pulls filings as they post, so the universe you're searching is the filed record, not last quarter's copy of it.
"What did this provision say in 2019, and when did it change?"
Traces tariff language through its history: what changed, when, and in which filing. Hours of expert document work compressed into a cited answer.
"What does that provision mean for our contracts?"
Connects your agreements and internal records to the filed tariffs, so the answer covers your exposure, not just the public text.
"Just tell me what controls and what it costs us."
The roll-up. Ask the commercial question in plain English and get the controlling provision, its history, and your position, with every source cited.
Built on the platform running in production at a Gulf Coast fabricator today. Quoting, margin, and shop operations, answered from your own job history.
"Do we have what these jobs need, and what's just sitting there?"
What's on hand, what's committed, and what's about to be short, reconciled against the jobs on the schedule instead of a count from last quarter.
"What should this job actually cost us to build?"
Builds estimates from your own history: labor, material, and the scope patterns that bit you last time. Evidence your estimator can defend in a quote review.
"What did jobs like this one actually take?"
Finds genuinely comparable past jobs, matched on structure and operational profile rather than keywords, with the actuals beside the quotes.
"Can the estimate read the drawings so we don't re-type them?"
Pulls scope from drawings into the estimate and the job record automatically, cutting the manual takeoff and the transcription errors that ride along with it.
"Where did we lose money last quarter, and why?"
Quoted versus actual, job by job, with the scope types, customers, and cost patterns that consistently erode margin surfaced in plain numbers.
"Where does time go between job start and delivery?"
Bottlenecks, routing, and throughput patterns pulled from your production data, so scheduling decisions run on evidence instead of instinct.
Whatever you call the things above, they're all assembled from the same tested building blocks. That's why deployments take weeks instead of quarters, and why everything behaves the same way: recommend, review, approve, log.
That's how everything on this page started. Tell us what's costing you time or margin. If it's worth solving, we'll scope it, build it, and argue later about whether it's an agent or a feature.