Generalist AI doesn't know your standards.
Ship repair is governed by thousands of interlocking standards — JFMM, NAVSEA Standard Items, AWS D1.1, MIL-STDs. YardOS: Shipyard Expert grounds your agent in the actual sources — current revision, cross-references, citations to the paragraph and page.
Private by default — every request runs under your account. Works with any MCP-capable client.
Why generalist AI fails in ship repair
Generic answers
Ask about surface preparation for exterior steel above the boottop and you'll get a plausible-sounding paragraph. It may not mention the correct NSI, table, or line — because the model was never trained on the current fiscal-year package.
Outdated information
Standard Items change on fiscal-year cycles. A model trained on public web data has no concept of "FY-28 versus FY-26." It will cite the wrong revision with the same confidence it cites the right one.
Uncited facts
Without a provenance trail, you cannot verify a claim against the controlling document. In ship repair, "I read it somewhere" is not good enough — the cognizant SUPERVISOR and the contract require the exact source.
Grounded in current source text
- JFMM Vol I–VII + Appendix 4-E
- NAVSEA Standard Items 105 items, FY-28 CH-1
- Standard Phraseology Uniform work-item language
- AWS D1.1 Structural welding code — steel
- MIL-STDs 170+ invoked standards
Every answer traces to a specific document, revision, and page. If we can't cite it, we don't assert it.
One hard question, answered from the source
A real agent session, condensed. The prompt, the tool trail, and the cited answer it returned. No paraphrasing from training data — every claim is traced to a source id and page.
- You
- “Use the naval-yard-expert MCP server. Do not answer from memory. For a surface ship work item involving exterior steel above the boottop, determine the applicable NAVSEA Standard Item, the specific table and line to use, the surface preparation requirement, the primer and coating system choices, and any important notes or caveats. Cite source ids and pages.”
- Agent · working
- find_sources “exterior steel above boottop” → yardos:nsi/FY-28/009-032 get 009-032 → “Cleaning and Painting Requirements; accomplish” · pp. 280–419 query_table #table-2 → Line 1, “Exterior surfaces above boottop” · cols A–G · pp. 348–371 get notes (2) (8) (33) (64) → caveats resolved · pp. 321–327
- Agent
-
The governing item is NSI 009-032, “Cleaning and Painting Requirements; accomplish.” For exterior steel above the boottop, use Table 2, “Steel Surfaces / Surface Ships,” Line 1 — vertical surfaces read from column G. Line 1 is one valid line in its row group; adjacent lines carry alternative coating systems. yardos:nsi/FY-28/009-032#table-2/line-1 · pp. 348–371
Surface preparation is near-white metal — NACE 2 / SSPC-SP 10 abrasive blast or WJ-2/M waterjetting. Spot-and-sweep may substitute with SUPERVISOR approval. …#table-2/line-1/col-A · note (33), p. 323
Coating system: primer MIL-PRF-23236 Type V/VI Class 5/7, 4–8 mils; a stripe coat of the same; topcoat Haze Gray MIL-PRF-24635 Type V/VI Class 2, 5–8 mils. The primer must share a manufacturer with the rest of the system; below 50 °F, low-temperature limits come from the NAVSEA-reviewed ASTM F718. …#table-2/line-1 · cols B, C, G · notes (64) p. 327, (8) p. 321
This package is authoritative for the served text, but the official NAVSEA publication and the cognizant SUPERVISOR remain controlling.
It answers the way an expert does
Name the governing standard, quote the relevant passage, and show where the answer came from.
- Orientation
- Start from the task at hand and find the controlling document fast.
- Passage lookup
- The governing paragraph, quoted, with its revision and page.
- Table facts
- Line-and-column values read straight from standards tables.
- Concept to source
- Ask in plain language; land on the exact item and section.
- Cross-references
- Follow invoked standards from document to document.
- Feedback
- Flag a gap or a citation for review without leaving the workflow.
Tiers gate speed, never accuracy
Every tier serves the same cited, grounded content. Lower tiers are rate-limited — they are never given degraded or uncited answers. Paying raises how fast and how much your agent can ask.
-
Free
$0
Where every account starts. Full search, paragraph-level citations, and feedback from the agent workflow.
20 requests/min
-
Individual
$3/mo
One person’s working tier — the same reads, at the pace of a full day’s work.
120 requests/min
-
Enterprise / Agent
Teams and agent fleets — pooled access for many agents under one agreement.
1,200 requests/min
Beyond Enterprise, an Unlimited + Raw tier adds unmetered rates and raw-data export rights.
Ground your agent in the sources it needs.
Create an account, connect your MCP client, and start asking against current shipyard standards.
Free tier — 20 requests/min, fully cited. Already registered? Sign in.