Tool Selection Guide: genai.mil vs Ask Sage

1. The Two-Tool World

The post-Jan-2026 landscape settled around two enterprise AI platforms. Both are first-class for the Marines today. Tool selection is itself a skill. Picking the wrong tool costs prompt iterations and review time even when the output is technically correct.

• genai.mil is the Marines' enterprise default chat platform (Jan 2026 MARADMIN, replaces NIPRGPT). Always-on conversational AI. CUI-authorized. Fast for single-shot prompts. Browser-based, no setup, no file uploads.
• Ask Sage is FedRAMP High and IL5 authorized in the default enclave; IL6 and Top Secret available through separate enclaves with separate provisioning. Azure Government or AWS GovCloud hosted (depending on the customer enclave). 150+ models including current Claude Opus and Sonnet, GPT-5 plus reasoning variants, Gemini 2.5, and Llama 4 Scout where available. Ingests DOCX, PDF, JSON, CSV, and code. API plus Continue.dev IDE integration. CUI and PHI authorized (DHA deployment Dec 2025).
• The Army runs Ask Sage on cArmy Cloud under a multi-year IDIQ. DHA uses it for clinical and operational workflows under a separate authorization. Adoption is past the experimental stage and the platform sits on real contracts; verify exact vehicle and ceiling against current contract data before quoting figures.

The post-Power-Platform pivot left a gap: Marines need real work done on real tenants, and the answer for any given task is one of these two tools. Picking the wrong one costs iterations.

2. Decision Tree

Read top to bottom. The first branch that matches your task is the answer. If two branches match, the lower one wins (it is more specific).

3. Quick Reference Table

Side-by-side capability comparison. Use this when you need to defend a tool choice to a peer or to a senior who asks "why not just use the other one."

4. Common Patterns

Five concrete workflows that show up across units. Each one is a worked example: scenario, tool, and why.

Pattern 1: Weekly readiness rollup

Use Ask Sage with Claude Opus 4.7. Upload last-week and this-week CSV exports. Ask for delta analysis: who fell off currency, who came on, what trends matter for the CO's brief. Why: dataset comparison is reasoning plus multi-file. genai.mil cannot ingest the CSVs and pasting them is lossy.

Pattern 2: Counseling statement first draft

Use genai.mil. One paragraph in (Marine's name, billet, period, key observations), one draft out. Why: single-shot. Ask Sage setup for a routine counseling is overkill and slower.

Pattern 3: Refactor a static-stack tool across three files

Use Ask Sage with a reasoning model. Upload the three files. Ask for the refactor in one pass. Why: multi-file context is required. A chat model loses track of file two by the time it reaches file three.

Pattern 4: High-stakes commander's brief

Use both. Run the same prompt through genai.mil and Ask Sage. Compare outputs. Reconcile differences. Why: cross-check protocol for high-stakes outputs (Course 4.5 Module 4). Disagreement between the two is a flag for human review. Agreement is corroboration.

Pattern 5: Quick regex for parsing T&R codes

Use genai.mil. Why: one-liner. Faster than Ask Sage's setup overhead. The whole interaction is under thirty seconds.

5. When You Only Have genai.mil (Tenant Limitations)

Not every Marine has Ask Sage access yet. On tenants where Ask Sage is not provisioned, Course 4.5 module work degrades gracefully. Here is how to ship anyway.

Workflow degradation

The Ask Sage phases of Course 4.5 modules degrade to paste-and-summarize on genai.mil. You lose multi-file context. You gain time on setup. The outputs are lower quality but they are still usable for low-stakes work.

Multi-file work in genai.mil only

Summarize each file in its own chat first. Be specific in the summary: keep names, dates, references, and any field that downstream prompts will need. Then start a fresh chat with the summaries as the context block. This is lossy but workable for two or three files. Past four files the loss compounds and the output is not defensible.

Dataset analysis in genai.mil only

Hand-aggregate the raw data to a 20-row summary in a textarea. Paste the summary. Analyze. Acceptable for small datasets where the aggregation step is itself the work. Useless for large datasets where the value is in the rows you would have missed.

File an access request

Ask Sage access is the unblock. Contact your unit's AI lead or the Marine Corps AI working group through the appropriate channel for your command. The request is routine. Document the gap in a Capability Gap Map row while you wait, so the access pathway gets shorter for the next Marine.

6. Picking the Right Ask Sage Model

Ask Sage exposes 150+ models. The decision in Section 2 picks the tool. This section picks the model within Ask Sage. Treating Ask Sage as monolithic costs iterations the same way picking the wrong tool does. The model choice is its own 201-level decision.

Model Switching Mid-Conversation

Ask Sage lets you switch models within a single conversation. Most Marines miss this. The pattern: Opus 4.7 for the Phase 1 architecture turn, Sonnet 4.6 for the Phase 2 implementation turns, Haiku 4.5 for the Phase 3 small-fix turns, then GPT-5 reasoning for the Phase 4 verification turn. You are using one model to QA another. Disagreement at the verification turn is a flag.

When Opus 4.7 Is Worth the Cost

Current Claude Opus is the heaviest model the tenant exposes. Reserve it for tasks where the reasoning premium pays for itself in fewer turns and better outputs: data-model proposals, multi-file refactors, architecture decisions, citation verification, the verification turn at the end of a long Sonnet or Haiku session. Do not default to Opus for tasks the workhorse Sonnet or a fast Haiku model would handle. Heavier reasoning models are slower and more quota-heavy; on bulk-seat tenants the budget pressure is real even when individual Marines do not see per-token billing.

7. The DoD / DoW Naming Note

Executive Order 14347 (5 September 2025) authorized "Department of War" as a secondary name and authorized officials including the Secretary of War to use that title in public communications and nonstatutory documents. The statutory name remains "Department of Defense" until Congress acts. Both names are in use today. Model training cutoffs predate the EO for most of the models Ask Sage exposes, so AI output defaults to "DoD." Marines drafting official documents should match the convention their command and the audience expect.

The Fix: A System-Prompt Snippet

Paste this at the start of any Ask Sage or genai.mil session that will produce official-document text where naming matters:

Where Naming Choice Matters Most

• Awards citations, FITREP comments, official correspondence (signed documents).
• SOPs, SOAs, and internal policy documents.
• Briefs for outside audiences (Congress, allies, civilian press) who may or may not be tracking the EO 14347 secondary-name authorization.

Where It Does Not Matter

• Drafting code or technical specifications that never reference the organization name.
• Internal informal correspondence where the audience knows the context.
• Citations to pre-EO source documents (cite them with their original organizational name as published).

8. CUI and OPSEC Anonymization Workflow

Both genai.mil and Ask Sage are CUI-authorized. Neither is PII-clean. Marines should scrub PII before pasting, even though the tenant allows it, because the discipline of scrubbing forces a second look at what is in the input. The discipline catches more spillage than tenant policy enforcement does.

The Pre-Paste Scrub Prompt

Run this on your input BEFORE pasting it into the working prompt. Paste your raw input where indicated, then paste the scrubbed output into your real prompt.

Review the following text for OPSEC, PII, and CUI markings. Produce a scrubbed version that is safe to paste into a CUI-authorized AI tool.

Rules:
- Replace specific Marine names with rank and billet (e.g., "Sgt Smith" becomes "the SNCO" or "Sgt A").
- Replace EDIPI with "EDIPI" or "{EDIPI}".
- Replace specific unit identifiers (UIC, PIIN, T-line) with the unit type ("an infantry battalion" instead of "1/4").
- Preserve T&R references, MCO citations, and authority references (those are public).
- Preserve dates and times unless they would compromise OPSEC.
- Flag anything that is not safe to scrub and should be removed entirely.

INPUT:
{PASTE_RAW_INPUT_HERE}

OUTPUT:
- The scrubbed version.
- A list of items you removed or flagged.

What to Scrub vs What to Keep

9. What Ask Sage Cannot Do (Yet)

Honest assessment. Saves Marines from wasting time on tasks Ask Sage cannot complete in the current tenant.

Connections Ask Sage Does Not Have

• GCSS-MC. No direct connector. You can paste exported reports or upload CSVs from GCSS-MC, but Ask Sage cannot pull live data.
• Marine Online (MOL). No connector. Same paste-and-summarize workflow as GCSS-MC.
• MarineNet. No connector. For PME or training-record queries, you upload the export or paste the text.
• DTS. No connector. Travel orders drafting is one-shot prompt territory.
• SharePoint write-back. Ask Sage can read uploaded files. It cannot push updates to a SharePoint list, library, or page.
• Microsoft Graph API. Not exposed in the tenant. Outlook, Teams, calendar workflows are out of scope.

Data Ask Sage Does Not Know About You

• Your unit's PIIN, UIC, T-line, or table of organization.
• Your unit's mission essential task list unless you provide it.
• Your unit's specific MCBul or unit-level standing orders.
• Any internal pubs that are not on the public Marine Corps Publications library.

Tasks Where Ask Sage Drifts

• Real-time information. Ask Sage models have training cutoffs. They do not know today's weather, today's MARADMIN, or today's roster changes.
• Math beyond two-step reasoning. Reasoning models help, but for serious math, use a spreadsheet.
• Long-document summarization where every word matters. Summary loses fidelity. Quote-back specific paragraphs if you need exactness.
• Citation verification. Ask Sage hallucinates MCO paragraph numbers frequently, even when it gets the MCO right. See the Citation Verification resource for the protocol.

10. Voice and Constraints

Print View

Use your browser Print function to produce a PDF for routing or for a unit AI binder. The page header chrome and the table of contents are hidden on print. Each section breaks cleanly.